Question 5

Explain the meaning of the following terms and give examples where appropriate:

a) Middleware[5]

b) Persistent / transient objects[5)

c) Logical clock[5]

d) Iterative / concurrent server[5]

e) Home-based approach to flat naming[5]

a) Middleware: Middleware refers to software that acts as a bridge between different systems or
applications, allowing them to communicate and interact with each other. It helps in managing and
facilitating communication, data exchange, and services between different systems. Examples of
middleware include web servers, application servers, message-oriented middleware, and database
middleware.

b) Persistent / transient objects:

- Persistent objects refer to objects that exist beyond the execution of a program and retain their state
even after the program terminates. These objects are usually stored in databases or files for long-term
storage.

- Transient objects, on the other hand, refer to objects that are temporary and exist only during the
execution of a program. They are not stored permanently and are typically used for temporary data
processing.

Example: In a banking application, customer information stored in a database would be considered as

persistent objects, while temporary variables used for calculations within the application would be
considered as transient objects.

c) Logical clock: A logical clock is a concept used in distributed systems to order events and provide a
partial ordering of events that occur in the system. It is not based on physical time but on logical
relationships between events. Each process in the system maintains its own logical clock to timestamp
events.

Example: In a distributed system, if two processes P1 and P2 communicate with each other, logical
clocks can be used to ensure that events are ordered correctly even if the physical clocks of the
processes are not synchronized.
d) Iterative / concurrent server:

- Iterative server: In an iterative server, the server processes requests from clients one at a time. It
handles one client request, completes it, and then moves on to the next client request. This approach is
simple but may not be efficient for handling multiple clients simultaneously.

- Concurrent server: In a concurrent server, the server can handle multiple client requests
simultaneously by creating a separate thread or process for each incoming request. This allows the
server to serve multiple clients concurrently, improving performance and responsiveness.

e) Home-based approach to flat naming:

In a home-based approach to flat naming, each object or resource is assigned a unique name based on
its location or home node in a distributed system. This approach simplifies naming and routing by
allowing each object to have a single flat name that includes the node where it resides.

Example: In a distributed file system using a home-based approach to flat naming, a file named
"data.txt" stored on a specific server might be accessed using a single flat name such as
"server1/data.txt". This naming convention helps in identifying the location of the file and routing
requests to the correct server.

Question 6

a) Describe a peer-to-peer architectural model for construction of a named distributed system of your
choice.

[8]

b) Describe the asymmetric cryptography technique and show how it can be used in supporting
security in distributed systems.

[8]

c) Explain synchronization with respect to distributed systems.

a) Peer-to-peer architectural model for construction of a named distributed system:

In a peer-to-peer architectural model, every node in the network is considered both a client and a
server, allowing for direct communication and resource sharing between nodes without the need for a
central server. This model is decentralized, scalable, and fault-tolerant.

To construct a named distributed system using a peer-to-peer architecture, nodes can be assigned
unique identifiers or names to facilitate communication and resource discovery. Each node maintains a
local routing table that maps keys or names to the corresponding nodes responsible for handling
requests related to those keys.

Example: In a peer-to-peer file-sharing system, nodes can be assigned names based on their IP
addresses or unique identifiers. When a node wants to download a file, it can query other nodes in the
network to locate the node that has the desired file based on its unique name or key.

b) Asymmetric cryptography technique and its use in supporting security in distributed systems:

Asymmetric cryptography, also known as public-key cryptography, uses a pair of keys (public key and
private key) for encryption and decryption. The public key is used for encryption and can be shared with
anyone, while the private key is kept secret and used for decryption. This technique is widely used in
supporting security in distributed systems for authentication, digital signatures, and secure
communication.

In distributed systems, asymmetric cryptography can be used for:

- Secure communication: Nodes can exchange encrypted messages using each other's public keys to
ensure confidentiality and integrity.

- Authentication: Nodes can verify the identity of each other by signing messages with their private keys
and verifying the signatures using the corresponding public keys.

- Digital signatures: Nodes can digitally sign messages to prove their authenticity and integrity, allowing
other nodes to verify the sender and detect tampering.

c) Synchronization in distributed systems:

Synchronization in distributed systems refers to the coordination of activities and data consistency
between multiple nodes or processes in a network. It is essential to ensure that different components of
the distributed system operate in a coordinated manner and maintain consistency across distributed
data.

In distributed systems, synchronization can be achieved through techniques such as:

- Distributed locking: Nodes use distributed lock management protocols to coordinate access to shared
resources and prevent conflicting updates.

- Clock synchronization: Nodes synchronize their clocks using clock synchronization algorithms to ensure
a consistent global time reference for ordering events.
- Distributed transactions: Nodes use distributed transaction protocols to ensure atomicity, consistency,
isolation, and durability (ACID properties) of transactions across multiple nodes.

Overall, synchronization in distributed systems is crucial for maintaining data consistency, ensuring
correctness of operations, and preventing race conditions and conflicts between nodes.

Question 1

Suppose you are running a large distributed on line shop with clients in different countries. To
minimize the costs associated with hosting your system, you decided to adopt a cloud based solution:

You rent a number of machines from a global cloud service provider. The provider's machines can host
arbitrary software that you supply. There are thousands of them, and they are located virtually all
over the world. Moreover, the provider supports dynamic on-demand acquisition and relinquishment
of machines, that is, at any time you can start additional instances of your software on arbitrary
provider's machines that are currently available or you can dispose of any no longer needed running
instances. Billing the additional resources is done automatically by the provider. Normally, you rent a
few machines in different parts of the world. However, occasionally the number of machines is
insufficient to serve your clients' requests. Typically this happens when certain shop items become
popular or before major holidays, but you are not always able to predict such situations. The great
majority of requests during such bursts are requests for browsing your shop and particular items.

Your goal is to minimize the costs incurred by hosting your system and not to lose elients due to a
degraded performance during request bursts.

a) Propose a logical organization of the sofiware comprising your system.

b) Propose a physical distribution of the software across the provider's machines.

c) Describe the mechanisms and algorithms you would employ to handle request bursts.

d) Discuss a few potential probions with the solutions from a)-c

a) Logical organization of the software comprising the system:

1. Frontend server: This server handles incoming requests from clients and distributes them to the
appropriate backend servers. It also serves static content for browsing the online shop.

2. Backend servers: These servers handle the actual processing of requests, such as retrieving product
information, processing orders, and handling user sessions.

3. Database server: This server stores product catalog data, user information, and order details.
4. Load balancer: This component distributes incoming requests among multiple backend servers to
ensure load balancing and high availability.

5. Caching server: This server caches frequently accessed data to reduce the load on the database server
and improve performance.

b) Physical distribution of the software across the provider's machines:

1. Deploy frontend servers in multiple regions close to the clients to minimize latency.

2. Deploy backend servers in locations with high demand to handle processing of requests efficiently.

3. Utilize the provider's dynamic on-demand acquisition and relinquishment of machines to scale up or
down based on demand.

4. Use a global load balancer to distribute incoming traffic among frontend and backend servers in
different regions.

5. Implement a distributed caching mechanism to store frequently accessed data close to the users.

c) Mechanisms and algorithms to handle request bursts:

1. Auto-scaling: Utilize auto-scaling capabilities provided by the cloud provider to automatically add or
remove machines based on traffic load.

2. Queueing: Implement a request queue to manage incoming requests during burst periods and ensure
fair processing.

3. Content delivery networks (CDNs): Cache static content on CDNs to reduce server load during high
traffic periods.

4. Dynamic load balancing: Use dynamic load balancing algorithms to distribute traffic efficiently among
available machines and maintain optimal performance.

5. Resource utilization monitoring: Implement monitoring tools to track resource usage and
performance metrics to make data-driven decisions on scaling and resource allocation.

d) Potential problems with the solutions:

1. Cost management: The automated scaling of resources can lead to unexpected costs if not properly
monitored and managed.
2. Data consistency: Distributing data across multiple locations can introduce challenges related to data
consistency and synchronization.

3. Security concerns: With a globally distributed system, maintaining security standards and compliance
across multiple regions can be complex.

4. Network latency: Distributing software across different regions can increase network latency,
affecting the overall performance of the system.

5. Dependency on cloud provider: The system's reliability is dependent on the cloud provider's
infrastructure and services, which can introduce risks in case of provider downtime or issues.

Question 2

2) Explain the fundamental models considered in design of DS.

b) Describe inter - process communication.

a) Fundamental models considered in the design of distributed systems:

1. Client-Server Model: In this model, clients request services or resources from servers, which respond
to these requests. It is a common model for distributing processing across multiple machines and
providing centralized access to resources.

2. Peer-to-Peer Model: In this model, all nodes in the network can act as both clients and servers,
allowing for direct communication and resource sharing between peers without the need for a central
server. It is a decentralized model that promotes scalability and fault tolerance.

3. Distributed Objects Model: In this model, objects are distributed across different nodes in the
network, and communication between objects is achieved through method invocation. It allows for a
more modular and flexible design of distributed systems.

4. Message-Passing Model: In this model, communication between nodes is achieved through message
passing, where messages are sent and received asynchronously. It provides a simple and efficient way
for nodes to communicate and exchange information.

b) Inter-process communication (IPC):

Inter-process communication refers to the mechanisms and techniques used for processes to
communicate and synchronize with each other in a distributed system. IPC is essential for coordinating
activities, sharing data, and achieving collaboration between different processes or components within a
distributed system. There are several methods of IPC, including:

1. Shared Memory: In shared memory IPC, multiple processes can access a common memory region that
is shared among them. Processes can read from and write to the shared memory segment, enabling fast
communication and data exchange.

2. Message Passing: Message passing IPC involves processes communicating by sending and receiving
messages. Messages can be of various types, such as synchronous or asynchronous, and can contain
data or commands. It provides a more flexible and secure way for processes to communicate compared
to shared memory.

3. Pipes: Pipes are a simple form of IPC that allows a unidirectional flow of data between two processes.
One process writes data to the pipe, and the other process reads from it. Pipes are commonly used for
communication between a parent process and its child process.

4. Sockets: Sockets are endpoints for communication between processes over a network. They can be
used for IPC between processes running on the same machine or on different machines. Sockets provide
a versatile and efficient way for processes to communicate over a network.

Overall, inter-process communication plays a crucial role in facilitating coordination, data sharing, and
collaboration between processes in a distributed system, enabling them to work together towards
achieving a common goal.

Question 3

a) Imagine you are a software engineer in a company that is planning a move towards using web-
based distributed computing. The technical manager has called for a discussion of the challenges
involved. The fact that the existing systems were originally designed with a view of being CPU-
efficient, causes her particular concern. She asks you to comment on whether her concerns are well-
founded.

Discuss whether you agree with your technical manager or not.

|131

b) Explain why synchronization becomes a more signiticant problem when the components of a
distributed system are very heterogeneous.

15]
c) Explain the importance of caching in a distributed system, clearly showing its importance in
avoiding inconsistent or stale data.

a) In the context of moving towards web-based distributed computing, the concern raised by the
technical manager regarding the existing systems being designed for CPU efficiency is well-founded.
When transitioning to a distributed system, there are several challenges that may arise, including:

1. Scalability: Web-based distributed systems often need to support a large number of users and handle
varying levels of traffic. The existing CPU-efficient design may not be optimized for handling distributed
workloads and scaling resources dynamically.

2. Network Communication: Distributed systems rely on network communication for inter-process

communication and data exchange. The design considerations for efficient CPU usage may not
necessarily align with the requirements for efficient network communication, which can impact overall
performance and latency.

3. Fault Tolerance: Distributed systems must be resilient to failures and errors that can occur in a
distributed environment. The existing system's focus on CPU efficiency may not have addressed fault
tolerance mechanisms such as redundancy, replication, and error handling.

4. Data Consistency: Ensuring consistency and coherence of data across distributed components can be
challenging. The existing system's design may not have considered the complexities of maintaining data
consistency in a distributed environment.

Therefore, it is important to reassess the design and architecture of the existing system to adapt it to the
requirements and challenges of web-based distributed computing.

b) Synchronization becomes a more significant problem in a distributed system with heterogeneous

components due to the following reasons:

1. Clock Synchronization: Heterogeneous components may have different clock speeds and time zones,
making it challenging to synchronize events and maintain a consistent notion of time across the system.

2. Communication Protocols: Different components may use varied communication protocols and
standards, leading to compatibility issues and the need for protocol conversion and translation.
3. Data Representation: Heterogeneous components may store and process data in different formats or
data structures, requiring transformation and mapping mechanisms for data interchange.

4. Resource Management: Synchronization of resources such as memory, disk space, and processing
power becomes complex in a heterogeneous environment, as each component may have different
capabilities and requirements.

Overall, the diverse nature of heterogeneous components in a distributed system introduces challenges
in coordinating and synchronizing activities, data, and resources efficiently.

c) Importance of caching in a distributed system to avoid inconsistent or stale data:

1. Performance Improvement: Caching frequently accessed data locally can reduce the latency of
retrieving data from remote servers, improving overall system performance and response times.

2. Data Consistency: Caching can help maintain data consistency by storing a copy of the data closer to
the requesting entity, reducing the likelihood of accessing outdated or stale data from remote sources.

3. Scalability: Caching can help distribute the load on backend servers by serving cached data locally,
reducing the burden on central servers and enabling horizontal scalability.

4. Resilience: In the event of network failures or downtime of remote servers, cached data can still be
accessed locally, ensuring continuous availability and functionality of the system.

5. Cost-Effectiveness: Caching reduces the need for frequent access to remote servers, resulting in cost
savings on bandwidth and resource utilization.

Overall, caching plays a crucial role in improving performance, ensuring data consistency, enhancing
scalability, and providing resilience in a distributed system, making it an essential component for
optimizing system efficiency and functionality.

Question 4
It is now common for people to make up their own holiday packages by booking flights, hotels and
excursions at different on line sites rather than choosing a pre-determined package offered by travel
agents. When booking these independently, however, a problem exists that makes it difficult to
guarantee that you get all the components you want for your holiday. For example, by the time you
finish booking your flight, it may be that the hotel you prefer no longer has any rooms, or that the
theatre event you must see has sold out. This leaves the unpleasant possibility of being left with a
flight to a city you no longer want to visit. A proposal has been put forward to set up an on line
business called allornothing.com to overcome this problem for users. The idea is that you select all the
components you want for your holiday on one single site, which then guarantees to keep them all on
hold until you come to make your final payment. If any one component fails to be booked for some
reason, the site guarantees that none of the components are booked / paid for. This way you end up
with nothing, or with exactly the bookings you want.

In the context of this proposal, discuss the following:

a) Identify issues relating to deadlock that exist in this system.

b) Explain how the issues identified in 4a above can be avoided.

151

c) Given that the booking process will need to operate over several web pages, describe how server
state can be preserved on the server from one page to the next.

15]

d) Describe how the system can be made robust in the face of users failing to complete the booking
process for whatever reason.

a) Issues relating to deadlock in the proposed system:

Deadlock can occur in the booking process of allornothing.com if multiple users attempt to book the
same components at the same time, leading to a situation where each user holds a resource required by
the other to proceed. In this scenario, if one user is waiting for a resource that is held by another user, a
deadlock can occur, preventing both users from completing their bookings.

b) Ways to avoid deadlock in the system:

1. Resource Allocation Strategy: Implement a resource allocation strategy that prevents deadlock, such
as utilizing a resource ordering policy where users must acquire resources in a specific order to avoid
circular waiting.
2. Timeout Mechanism: Implement a timeout mechanism that releases resources held by a user if they
do not complete the booking process within a specified time limit, preventing resources from being
unnecessarily locked.

3. Deadlock Detection: Implement a deadlock detection algorithm that periodically checks for deadlock
conditions and resolves them by rolling back transactions or restarting the booking process.

4. System Design: Design the system in a way that minimizes the likelihood of deadlock, such as limiting
the number of resources that can be held simultaneously or ensuring that resources are released
promptly after use.

c) Preserving server state across web pages:

To preserve server state across multiple web pages during the booking process, the system can utilize
techniques such as:

1. Session Management: Utilize session variables to store user-specific information and state across
multiple web pages. Sessions can be maintained using cookies, URL rewriting, or hidden form variables.

2. Hidden Form Fields: Store important data in hidden form fields within HTML forms, allowing the
server to retrieve and update the data as the user navigates through different pages.

3. Server-side State Management: Use server-side databases or files to store and retrieve session data,
allowing the server to maintain state information even when the user navigates away from the current
page.

d) Ensuring system robustness in the face of users failing to complete the booking process:

1. Transaction Rollback: Implement a transaction rollback mechanism that cancels bookings and releases
resources if the user fails to complete the booking process or times out. This prevents incomplete
bookings from affecting system availability.

2. Error Handling: Provide clear error messages and prompts to guide users through the booking
process, informing them of any issues or incomplete steps that need to be addressed.
3. Data Validation: Validate user input at each step of the booking process to ensure that only valid and
complete information is submitted, reducing the likelihood of errors or incomplete bookings.

4. Booking Reservations: Reserve booked components for a limited time before final confirmation and
payment, allowing for temporary holds on resources to prevent conflicts and ensure availability for
other users

Question 1

A shared database uses two-phase commit to ensure that either all shard servers commit their part of
each transaction, or none of them do. A database client executing a transaction sends the
transaction's puts and gets to the shard servers, and then uses a transaction coordinator (TC) to
execute two-phase commit for the transaction. As a reminder, the steps of two-phase commit are as
follows:

A. The TC sends a PREPARE message to each participant (each shard server that is involved in the
transaction).

B. Each participant replies with YES or NO, according to whether the participant is able to commit.

C. If all participants answer YES, the TC sends a COMMIT message to each participant. If any
participant answers NO, or if the TC times out while waiting for replies, the TC sends out ABORT
messages.

D. If a participant receives a COMMIT message, it makes its part of the transaction's updates
permanent, and releases locks. If a participant receives an ABORT message, it forgets about the
transaction's updates, and releases locks.

Suppose a TC sends a COMMIT message to one of a transaction's multiple participants, but the TC
crashes before sending any more messages.

a) Explain what the TC must do after it reboots.

[10]

b) Explain why it would not be OK for the client to find a new coordinator, TC2, and ask TC2 to run
two-phase commit again for the transaction (starting with the

PREPARE messages). It is sufficient to supply an example scenario that leads to an incorrect outcome..

a) After the TC reboots following a crash, it must perform recovery actions to ensure the integrity of the
transaction. The following steps can be taken by the TC after rebooting:
1. Checkpointing: The TC should have a mechanism to checkpoint the state of the transaction before the
crash. This checkpoint can help the TC resume the transaction from where it left off.

2. Contact Participants: The TC should first contact the participants that it had previously sent the
COMMIT message to. It needs to ensure that the participants have not made their updates permanent
based on the COMMIT message.

3. Resend Messages: If the participants have not committed their updates, the TC needs to resend the
COMMIT message to all participants to complete the transaction. It should ensure all participants are in
sync before proceeding with the COMMIT phase.

4. Timeout Handling: The TC should handle timeout scenarios carefully to avoid inconsistencies and
ensure that all participants are properly coordinated.

5. Data Recovery: If necessary, the TC should handle data recovery and ensure that any changes made
by participants are consistent and durable.

b) It would not be appropriate for the client to find a new coordinator, TC2, and ask TC2 to run two-
phase commit again for the transaction because doing so could lead to an incorrect outcome. Here is an
example scenario to illustrate why this approach is not advisable:

Scenario:

1. Initially, TC sends the PREPARE message to all participants, and all participants respond with YES,
indicating they are ready to commit.

2. However, before the TC can send the COMMIT message to all participants, it crashes.

3. The client, unaware of the crash, decides to find a new coordinator, TC2, and asks TC2 to run two-
phase commit again for the transaction.

4. TC2 sends a new PREPARE message to all participants, and one of the participants, Participant A,
replies with NO this time, while other participants respond with YES.

5. Based on the new responses, TC2 decides to abort the transaction and sends ABORT messages to all
participants.

6. Participant A, which had previously received a COMMIT message from the first TC, has already made
its updates permanent based on that message. Receiving an ABORT message from TC2 now contradicts
the previous decision made during the first execution of two-phase commit.
This scenario demonstrates the potential inconsistency that can arise if a new coordinator is introduced
to run two-phase commit again without proper coordination and awareness of the previous transaction
state. It is crucial to ensure that the transaction coordinator maintains consistency and integrity
throughout the entire transaction process.

Question 2

It is now common for people to make up their own holiday packages by booking flights, hotels and
excursions at different online sites rather than choosing a pre-determined package offered by travel
agents. When booking these independently, however, a problem exists that makes it difficult to
guarantee that you get all the components you want for your holiday.

For example, by the time you finish booking your flight, it may be that the hotel you prefer no longer
has any rooms, or that the theatre event you must see has sold out. This leaves the unpleasant
possibility of being left with a flight to a city you no longer want to visit. A proposal has been put
forward to set up an online business called allornothing.com to overcome this problem for users. The
idea is that you select all the components you want for your holiday on one single site, which then
guarantees to keep them all on hold until you come to make your final payment. If any one
component fails to be booked for some reason,

Page 2 of 4

the site guarantees that none of the components are booked / paid for. This way you end up with
nothing, or with exactly the bookings you want.

In the context of this proposal, discuss the following:

a) Identify issues relating to 'security' that exist in this system.

[10)

b) Explain how the issues identified in 2a above can be avoided.

[5]

c) Describe how the system can be made robust in the face of users failing to complete the booking
process for whatever reason.

[10)

a) Issues relating to security in the proposed system:

1. Data Privacy: When users provide personal and financial information to make bookings on the
website, there is a risk of this data being compromised if the website's security measures are not robust.
This could lead to identity theft, fraud, or unauthorized access to sensitive information.

2. Payment Security: Since users will be making payments for their holiday packages through the
website, there is a concern about the security of payment transactions. If the payment gateway is not
secure, users' financial information could be intercepted by malicious actors.

3. Data Integrity: There is a risk of data manipulation or corruption if the system is not properly secured.
Unauthorized parties could potentially modify bookings, alter prices, or delete critical information.

b) Ways to avoid security issues:

1. Secure Socket Layer (SSL) Encryption: Implement SSL certificates to encrypt data transmitted between
users and the website, ensuring that information is secure during transit.

2. Secure Payment Gateway: Use trusted and secure payment gateways that comply with Payment Card
Industry Data Security Standard (PCI DSS) to protect users' financial information during transactions.

3. User Authentication: Implement secure login mechanisms, such as multi-factor authentication, to

verify the identity of users accessing the website and prevent unauthorized access.

4. Data Encryption: Encrypt sensitive data stored in databases to prevent unauthorized access and
ensure data confidentiality.

5. Regular Security Audits: Conduct regular security audits and penetration testing to identify and
address vulnerabilities in the system proactively.

c) Making the system robust in the face of users failing to complete the booking process:

1. Booking Timeout: Implement a booking timeout mechanism to release held bookings if users fail to
complete the transaction within a specified time period, ensuring availability for other users.

2. Error Handling: Provide clear feedback and error messages to users during the booking process to
guide them in completing the transaction successfully.
3. Reservation System: Implement a reservation system that temporarily holds bookings until final
payment is made, allowing users some flexibility in completing the booking process without losing their
desired components.

4. Customer Support: Offer responsive customer support to assist users who may encounter difficulties
during the booking process, helping them to complete their transactions successfully.

5. Transaction Logging: Maintain transaction logs to track the progress of bookings and identify any
incomplete transactions, allowing for follow-up and resolution with the users.

Question 3

(a) When distributed systems are designed and engineered, certain fundamental properties have to
be taken into account, including:

1. Concurrent execution of components

2. Independent failure modes

3. No global time

Give three examples of the implications of these properties (separately or in combination) on the
engineering of large-scale, widely distributed systems.[12)

(b) (i) Define role-based access control (RBAC).(2]

ii) Outline how RBAC could be used for a national healtheare system comprising many admipistration
domains such as primary care practices, hospitals, specialist clinics, etc.

Principals may, from time to time, work in domains other than their home domain, and must be
authorised to do so.[3)

(jii) A national Electronic Health Record (EHR) service must be accessible from all domains.

It is required by law that access control policy should be able to capture exclusion of principals and
relationships between them. How could this requirement be met in an RBAC design?[8]

(a)

1. Concurrent execution of components:

- Implication: In a distributed system where components can execute concurrently, engineers need to
consider issues related to synchronization, consistency, and concurrency control. This includes ensuring
that shared resources are accessed in a coordinated manner to prevent data corruption and maintaining
proper synchronization mechanisms to handle concurrent requests effectively.

- Example: In a large-scale distributed system where multiple users are accessing and updating a shared
database concurrently, engineers need to implement mechanisms like locks, transactions, and isolation
levels to maintain data consistency and avoid conflicts.

2. Independent failure modes:

- Implication: In a distributed system where components can fail independently, engineers must design
fault-tolerant mechanisms to handle failures gracefully without compromising the system's availability
and reliability. This includes implementing redundancy, replication, and error detection and recovery
strategies to mitigate the impact of component failures.

- Example: In a widely distributed system where nodes can fail independently, engineers can use
techniques like replication and load balancing to ensure that service availability is not affected if one or
more components fail.

3. No global time:

- Implication: In a distributed system where there is no global time reference, engineers need to address
challenges related to event ordering, synchronization, and ensuring consistency across distributed
components. This includes implementing protocols for clock synchronization, handling causality
constraints, and managing distributed transactions effectively.

- Example: In a large-scale distributed system where timestamps are used to determine the order of
events, engineers need to carefully design algorithms and protocols to maintain causality and ensure
that events are processed in the correct order.

(b)

(i) Role-Based Access Control (RBAC) is a security model that restricts system access to authorized users
based on their roles within an organization. Each role is assigned specific permissions, and users are
granted access based on their role, rather than individual user identities.

(ii) In a national healthcare system comprising multiple administration domains, RBAC can be used to
manage access control effectively. Principals, such as healthcare professionals, can be assigned roles
based on their responsibilities and access requirements within different domains. For example:

- A doctor working in a primary care practice may have a "Primary Care Physician" role, granting access
to patient records and treatment plans within that domain.
 - If the same doctor needs to work in a hospital or specialist clinic, they can be assigned additional roles,
such as "Hospital Physician" or "Specialist Consultant," to access relevant information and systems in
those domains.

(iii) To meet the requirement of capturing exclusions of principals and relationships between them in an
RBAC design for a national Electronic Health Record (EHR) service, engineers can:

- Implement hierarchical roles: Define roles with varying levels of access privileges and relationships
between roles to capture complex authorization requirements.

- Use constraints and role hierarchies: Establish exclusion rules and constraints to restrict access to
sensitive information based on relationships between principals, ensuring that only authorized users can
access certain data.

- Role activation and deactivation: Allow administrators to activate or deactivate roles for principals
based on their status, responsibilities, and affiliations within different domains, ensuring that access
rights are aligned with their current roles and permissions.

Question 4

a) In most distributed applications, message exchange is a means to an end. For example, messages
between the client and an e-commerce server carry information that allows a transaction to be
completed (or not) but the content is not the main purpose of that exchange: the main purpose is to
buy a product or a service. Identify and explain two examples of kinds of applications that run in
smartphones in which message exchange is the main purpose. [10]

b) Draw two diagrams whose purpose is to illustrate the essential difference between

threaded and distributed computations.

[6]

Page 3 of 4

c) Assume a modern desktop computer running a modern operating system. Assume the singh user
currently logged in to this computer launches in quick sequence such graphical user interface-based
applications as a word processor and a presentation program.

i) Explain whether or not, the given scenario is likely to involve concurrent computing.

iii) Explain whether or not, the given scenario is likely to involve distributed computing.

a) Examples of Smartphone Applications where Message Exchange is the Main

Purpose:

1. Messaging Apps: One of the most common examples where message exchange is the
primary function is messaging applications like WhatsApp, Facebook Messenger, or
 Signal. These apps are designed specifically for users to exchange messages,
multimedia content, and engage in real-time communication with individuals or groups.

2. Push Notification Services: Another example is push notification services like those
used by social media platforms, news apps, or email clients. These services send
notifications to users’ smartphones based on specific triggers or events, and the main
purpose is to deliver timely information or updates directly to the user’s device.

b) Diagrams Illustrating Threaded and Distributed Computations:

1. Threaded Computation Diagram:

 [Diagram illustrating threaded computation]

2. Distributed Computation Diagram:

 [Diagram illustrating distributed computation]

c) Scenario Involving Concurrent and Distributed Computing:

i) Concurrent Computing Involvement:

 The scenario described involving launching a word processor and a presentation

program in quick succession on a modern desktop computer is likely to involve
concurrent computing. Concurrent computing refers to the ability of a system to execute
multiple tasks simultaneously. In this case, both applications can run concurrently,
allowing the user to work on both tasks at the same time.

ii) Distributed Computing Involvement:

 The scenario described is not likely to involve distributed computing. Distributed

computing involves multiple interconnected systems working together on a task, sharing
resources and processing capabilities. In this scenario, the applications are running on
a single desktop computer and are not distributed across different machines.

Question 5

Write notes on the Byzantine Generals problem and offer solutions.[25]

 Byzantine Generals Problem:

 The Byzantine Generals Problem is a game theory problem that illustrates the
challenges decentralized parties face in reaching consensus without a trusted central
authority.
 In this scenario, multiple generals surround a city and must decide on a coordinated
attack plan. However, they lack secure communication channels and must contend with
the presence of traitorous generals who may alter or intercept messages.
 The essence of the problem lies in how loyal generals can come to an agreement
despite the potential interference from traitors, highlighting the need for a robust
consensus mechanism.

Solutions to the Byzantine Generals Problem:

1. Blockchain Technology:

 Blockchain technology, as exemplified by Bitcoin, offers a solution to the Byzantine

Generals Problem by providing a decentralized, transparent, and trustless system for
verifying transactions.
 Through a distributed ledger maintained by network nodes, blockchain ensures that all
participants can agree on the validity of transactions without relying on a central
authority.
 The use of cryptographic techniques and consensus mechanisms like Proof of Work
enhances security and prevents tampering with the shared ledger.

2. Byzantine Fault Tolerance (BFT):

 Byzantine Fault Tolerance is another approach to addressing the Byzantine Generals

Problem by enabling systems to function correctly even in the presence of faulty or
malicious components.
 BFT protocols ensure that a distributed network can reach consensus despite potential
Byzantine failures, enhancing reliability and fault tolerance.

3. Consensus Mechanisms:

 Various consensus mechanisms such as Proof of Work (PoW), Proof of Stake (PoS),
and Practical Byzantine Fault Tolerance (PBFT) play a crucial role in mitigating the risks
posed by the Byzantine Generals Problem.
 These mechanisms establish rules for validating transactions, achieving agreement
among network participants, and maintaining the integrity of the system.
Or
The Byzantine Generals problem is a classic computer science problem that deals with achieving
consensus among a group of distributed entities when some of these entities may be faulty or malicious.
The problem is named after the Byzantine Generals who faced a similar scenario where they needed to
reach an agreement on a coordinated attack or retreat, but some of the generals could be traitors
sending conflicting messages.

In the context of distributed systems, the Byzantine Generals problem can be stated as follows: a group
of generals (nodes) must come to an agreement on a common decision, such as attacking or retreating,
despite the presence of faulty or malicious nodes that may send conflicting or incorrect information. The
challenge is to reach consensus in the presence of Byzantine failures, where faulty nodes may act
arbitrarily and deceive other nodes.

Here are some notes on the Byzantine Generals problem and potential solutions:

1. Challenges of the Byzantine Generals Problem:

- Byzantine failures: In a distributed system, nodes may fail in arbitrary ways, including sending
contradictory or incorrect messages.

- Lack of central authority: There is no central node that can be trusted to provide correct information or
make decisions for the entire system.

- Asynchronous communication: Nodes communicate over an unreliable network where messages can
be delayed, lost, or corrupted.

2. Solutions to the Byzantine Generals Problem:

- Byzantine Fault-Tolerant (BFT) algorithms: Byzantine fault-tolerant algorithms are designed to achieve
consensus in the presence of Byzantine failures. These algorithms typically involve redundancy,
replication, and cryptographic techniques to tolerate Byzantine faults.

- Practical Byzantine Fault Tolerance (PBFT): PBFT is a popular Byzantine fault-tolerant algorithm that
uses a consensus protocol to reach an agreement among nodes. PBFT ensures safety and liveness
properties even in the presence of Byzantine faults.

- Proof of Stake (PoS) and Proof of Work (PoW): Blockchain consensus algorithms like PoS (used in
Ethereum 2.0) and PoW (used in Bitcoin) provide solutions to the Byzantine Generals problem by
leveraging economic incentives and cryptographic mechanisms to achieve consensus without a trusted
central authority.

- Hashgraph: Hashgraph is a distributed ledger technology that offers a Byzantine fault-tolerant

consensus algorithm based on gossip about gossip and virtual voting.
- Practical applications: The Byzantine Generals problem has practical applications in distributed
databases, blockchain networks, decentralized consensus protocols, and secure multi-party
computations.

In conclusion, the Byzantine Generals problem highlights the challenges of achieving consensus in
distributed systems with faulty or malicious nodes. By employing Byzantine fault-tolerant algorithms,
cryptography, and decentralized consensus mechanisms, it is possible to address the Byzantine Generals
problem and ensure the integrity and reliability of distributed systems in the face of Byzantine failures.

Question 6

Describe how LLMs such as the recently launched Open Als GPT 4 exhibit the true characteristics that
define Distributed Systems, highlighting the key issues the face and how they resolve them.[25]

LLMs (Large Language Models) like the recently launched OpenAI's GPT-4 exemplify the true
characteristics that define Distributed Systems by showcasing massive scale, distributed computation,
fault tolerance, and decentralized decision-making. These characteristics present key challenges that
LLMs face, and innovative solutions are employed to address them effectively.

1. Massive Scale:

LLMs like GPT-4 operate on a massive scale, processing vast amounts of data and performing complex
computations. To handle this scale, distributed systems are used to distribute the workload across
multiple nodes or servers. This allows for parallel processing, reducing the time required for training and
inference.

2. Distributed Computation:

Distributed computation is essential for LLMs like GPT-4, as training and running such large models
require distributed processing. By partitioning the model across different nodes and using parallel
computation, LLMs can handle the immense amount of data and calculations efficiently.

3. Fault Tolerance:

One of the key issues that LLMs face is the potential for system failures or errors during computation.
Distributed systems implement fault tolerance mechanisms to ensure that the system can continue
operating even if individual nodes fail. Techniques such as replication, redundancy, and checkpointing
are used to recover from failures and maintain system reliability.

4. Decentralized Decision-Making:
LLMs like GPT-4 rely on decentralized decision-making processes, where different nodes collaborate to
reach a consensus on the next steps in the computation. Distributed systems utilize consensus
algorithms, such as the Byzantine fault-tolerant protocols, to ensure that all nodes agree on the
decisions being made, even in the presence of faulty or malicious nodes.

5. Key Issues Faced by LLMs and Solutions:

a) Data Processing Speed: LLMs require high-speed data processing to train and run inference efficiently.
Distributed systems use parallel processing and distributed computing to speed up data processing and
reduce latency.

b) Scalability: As the size of LLMs grows, scalability becomes a challenge. Distributed systems can scale
horizontally by adding more nodes to the system, allowing LLMs to handle larger models and datasets.

c) Resource Management: LLMs require efficient resource management to optimize performance.

Distributed systems use resource allocation algorithms and load balancing techniques to distribute
resources effectively and prevent bottlenecks.

d) Communication Overhead: Communication overhead in distributed systems can impact performance.

Techniques such as message passing optimization, data compression, and efficient network protocols
are used to minimize communication overhead in LLMs.

In conclusion, LLMs like DeepAI's GPT-4 exemplify the characteristics of Distributed Systems by
leveraging distributed computation, fault tolerance, scalability, and decentralized decision-making. By
addressing key challenges through innovative solutions, LLMs can overcome the complexities of large-
scale language modeling and deliver high-performance natural language processing capabilities.

Question 1

A shared database uses two-phase commit to ensure that either all shard servers commit their part of
each transaction, or none of them do. A database client executing a transaction sends the
transaction's puts and gets to the shard servers, and then uses a transaction coordinator (TC) to
execute two-phase commit for the transaction. As a reminder, the steps of two-phase commit are as
follows:

A. The TC sends a PREPARE message to each participant (each shard server that is involved in the
transaction).

B. Each participant replies with YES or NO, according to whether the participant is able to commit.

C. If all participants answer YES, the TC sends a COMMIT message to each participant. If any
participant answers NO, or if the TC times out while waiting for replies, the TC sends out ABORT
messages.
D. If a participant receives a COMMIT message, it makes its part of the transaction's updates
permanent, and releases locks. If a participant receives an ABORT message, it forgets about the
transaction's updates, and releases locks.

Suppose a TC sends a COMMIT message to one of a transaction's multiple participants, but the TC
crashes before sending any more messages.

a) Explain what the TC must do after it reboots.

[10)

b) Explain why it would not be OK for the client to find a new coordinator, TC2, and ask

TC2 to run two-phase commit again for the transaction (starting with the PREPARE messages), It is
sufficient to supply an example scenario that leads to an incorrect outcome.

[15]

a) After the TC reboots following a crash before completing the two-phase commit protocol for a
transaction, it must perform the necessary recovery steps to ensure the consistency and integrity of the
transaction. The TC can follow the following actions:

1. When the TC reboots, it should first check the status of the previous transaction that failed to
complete due to the crash. It needs to determine whether any COMMIT messages were sent to the
participants before the crash.

2. If any COMMIT messages were sent but not received by all participants, the TC should initiate a
recovery procedure. It can communicate with the participants to check if they received the COMMIT
message and acted upon it.

3. If all participants have not received the COMMIT message or have not committed their parts of the
transaction, the TC should abort the transaction by sending an ABORT message to all participants. This
will ensure that the transaction is rolled back and the system returns to a consistent state.

b) It would not be acceptable for the client to find a new coordinator, TC2, and ask TC2 to run the two-
phase commit again for the transaction because this could lead to an incorrect outcome due to the
following scenario:

Imagine a situation where the original TC had sent a COMMIT message to one of the participants in the
transaction before crashing. The participant that received the COMMIT message may have already
committed its part of the transaction and made the updates permanent in the database.

If the client were to engage a new coordinator, TC2, and restart the two-phase commit process from the
beginning, including sending PREPARE messages again, the previously committed participant may
receive a PREPARE message from TC2. Since the participant has already committed its part of the
transaction and made the updates permanent, it may reply with a NO instead of YES in response to the
PREPARE message.
This would result in an inconsistent state where some participants have committed the transaction,
while others have not. This violates the atomicity property of transactions, where all participants should
either commit or abort the transaction together to maintain data consistency.

Therefore, it is crucial for the TC to handle the recovery process correctly and maintain the integrity of
the transaction to prevent such inconsistencies in the database.

Question 1

Suppose you are running a large distributed on line shop with clients in different countries. To
minimize the costs associated with hosting your sysiom, you decided to adopt a cloud based solution:

You rent a number of machines from a global cload service provider. The provider's machines can host
arbitrary software that you supply. There are thousands of them, and they are located virtually all
over the world. Moreover, the provider supports dynamic on-demand acquisition and relinquishment
of machines, that is, at any time you can start additional instances of your software on arbitrary
provider's machines that are currently available or you can dispose of any no longer needed running
instances. Billing the additional resources is done automatically by the provider. Normally, you rent a
few machines in different parts of the world. However, occasionally the number of machines is
insufficient to serve your clients' requests. Typically this happens when certain shop items become
popular or before major holidays, but you are not always able to predict such situations. The great
majority of requests during such bursts are requests for browsing your shop and particular items.

Your goal is to minimize the costs incurred by hosting your system and not to lose clients due to a
degraded performance during request bursts.

a) Propose a logical organization of the software comprising your system.

b) Propose a physical distribution of the software across the provider's machines.

c) Describe the mechanisms and algorithms you would employ to handle request bursts.

d) Discuss a few potential problems with the solutions from a)-c).

a) Logical Organization of the Software:

1. Frontend Servers: Handle client requests, serve web pages, and process user interactions.

2. Backend Servers: Manage inventory, process orders, and handle databases.

3. Load Balancer: Distribute incoming traffic across multiple frontend servers for load balancing and high
availability.

4. Caching Mechanism: Cache frequently accessed data to reduce database load.

5. Queueing System: Queue requests during bursts to prevent overwhelming the system.

6. Monitoring System: Monitor system performance and resource usage.

b) Physical Distribution of the Software:

1. Deploy frontend servers in various regions to reduce latency for clients.

2. Backend servers can be distributed for redundancy and data locality.

3. Use cloud provider's regions to ensure availability and scalability.

4. Deploy load balancers in front of frontend servers to distribute traffic efficiently.

c) Mechanisms and Algorithms for Handling Request Bursts:

1. Auto-Scaling: Automatically scale resources based on predefined thresholds or demand spikes.

2. Horizontal Scaling: Add more instances of software across different provider machines during bursts.

3. CDN (Content Delivery Network): Serve static content from edge servers closer to clients.

4. Caching Strategies: Implement caching mechanisms to reduce database load and serve static content
quickly.

5. Request Throttling: Limit the number of requests accepted during bursts to prevent system overload.

d) Potential Problems with the Solutions:

1. Cost Management: Auto-scaling and dynamic resource allocation could lead to unexpected cost
increases if not properly managed.

2. Data Consistency: Maintaining data consistency across distributed systems can be challenging,
especially during bursts.

3. Performance Degradation: Load balancing algorithms may not always distribute traffic efficiently,
leading to performance issues.

4. Security Concerns: Distributing software across multiple machines increases the attack surface,
requiring robust security measures.

5. Resource Wastage: Instances may be running unnecessarily during off-peak hours, leading to resource
wastage and increased costs.

question

a) Explain the fundamental models considered in design of DS.

b) Describe inter - process communication.

a) Fundamental Models in the Design of Distributed Systems:

1. Client-Server Model: In this model, the system is divided into clients that request services or
resources and servers that provide those services or resources. Clients initiate requests, and servers
respond to them, enabling a clear separation of concerns and scalability in distributed systems.

2. Peer-to-Peer Model: In a peer-to-peer model, all nodes in the system have equal responsibilities and
capabilities. Nodes can act as both clients and servers, sharing resources and information directly with
one another without the need for a central server.

3. Hierarchical Model: The hierarchical model organizes nodes in a tree-like structure, with higher-level
nodes managing and coordinating lower-level nodes. This model offers scalability, fault tolerance, and
easier management of the distributed system.

4. Distributed Objects Model: In this model, distributed components communicate by invoking methods
on remote objects. Objects encapsulate both data and behavior, allowing for easy distribution and
communication between different components in the system.

5. Event-Driven Model: In an event-driven model, components in the distributed system communicate

through events and messages. Components can publish events, subscribe to events, and react to events,
enabling loosely coupled and scalable systems.

b) Inter-Process Communication (IPC):

Inter-process communication refers to the mechanisms and techniques used by processes running on
different nodes or machines to communicate with each other in a distributed system. IPC allows
processes to exchange data, synchronize activities, and coordinate their operations. Some common
methods of IPC include:

1. Message Passing: Processes communicate by sending messages to each other. Messages can be
synchronous or asynchronous and can contain data or control information.

2. Remote Procedure Call (RPC): RPC allows a process to execute a procedure or method on a remote
machine as if it were a local procedure call. The calling process sends a request to the remote process,
which executes the procedure and returns the result.

3. Shared Memory: Processes can communicate by sharing a common memory segment. Changes made
by one process to the shared memory are visible to other processes, enabling fast and efficient
communication.

4. Sockets: Processes can communicate over a network using socket programming. Sockets provide an
interface for processes to send and receive data over a network, enabling communication between
processes running on different machines.

Using IPC mechanisms, processes in a distributed system can communicate, coordinate their actions,
share resources, and work together to achieve common goals. Effective IPC implementation is essential
for designing reliable, responsive, and scalable distributed systems.
Question 3

a) Imagine you are a software engincer in a company that is planning a move towards using web-
based distributed computing. The technical manager has called for a discussion of the challenges
involved. The fact that the existing systems were originally designed with a view of being CPU-
efficient, causes her particular concern. She asks you to comment on whether her concerns are well-
founded,

Discuss whether you agree with your technical manager or not

b) Explain why synchronization becomes a more significant problem when the components of a
distributed system are very heterogencous.

5) Explain the importance of caching in a distributed system. clearly showing its importance in
avoiding inconsistent or stale data

According to www.iAsk.ai Ask Ai Questions Search Engine:

a) Response:

I agree with the technical manager’s concerns about the challenges involved in moving towards
web-based distributed computing when the existing systems were originally designed with a
focus on CPU efficiency. Transitioning to a distributed system from a CPU-efficient design can
pose several challenges:

1. Resource Allocation: In a distributed system, resources are spread across multiple

machines, which can lead to complexities in resource allocation and management. The
shift from a centralized CPU-efficient system to a distributed model may require
rethinking resource utilization strategies.
2. Network Overhead: Distributed systems rely heavily on network communication for
coordination and data exchange. The transition may introduce additional network
overhead compared to the more localized processing in a CPU-efficient system.
3. Scalability Concerns: Distributed systems need to scale horizontally by adding more
machines to handle increased loads. The existing CPU-efficient design may not have
considered this scalability aspect, requiring careful planning to ensure seamless
scalability in the new architecture.
4. Fault Tolerance: Distributed systems must be resilient to failures in individual
components or machines. Adapting from a CPU-efficient design to distributed computing
involves implementing fault-tolerant mechanisms to maintain system reliability.

In conclusion, the technical manager’s concerns are well-founded, and addressing these
challenges will be crucial for a successful transition to web-based distributed computing.
b) Synchronization Challenges in Heterogeneous Distributed Systems:

Synchronization becomes a more significant problem in heterogeneous distributed systems due

to the following reasons:

1. Diverse Technologies: Heterogeneous systems consist of components running on

different platforms, using various technologies and protocols. Ensuring synchronization
across these diverse components requires handling compatibility issues and differing
synchronization mechanisms.
2. Communication Overhead: Synchronizing data and processes across heterogeneous
components often involves increased communication overhead. Coordinating actions
between components with varying processing speeds or communication protocols can
lead to delays and inefficiencies.
3. Consistency Maintenance: Maintaining data consistency becomes complex when
dealing with heterogeneous components that may interpret data or timestamps differently.
Ensuring consistent views of shared data across diverse systems requires robust
synchronization mechanisms.

Addressing synchronization challenges in heterogeneous distributed systems necessitates careful

design considerations, including standardized communication protocols, efficient data sharing
mechanisms, and synchronization strategies tailored to accommodate the diversity of the system.

c) Importance of Caching in Distributed Systems:

Caching plays a vital role in distributed systems by enhancing performance and avoiding
inconsistent or stale data:

1. Improved Latency: Caching frequently accessed data locally reduces the need to fetch it
from remote sources, significantly improving response times and reducing latency in
distributed environments.
2. Data Consistency: By caching data at strategic points within the system, inconsistencies
due to delays in updating shared data can be mitigated. Caching helps maintain data
coherence by serving up-to-date information when needed.
3. Load Balancing: Caching can distribute load efficiently by reducing repeated requests
for the same data across multiple components. This helps prevent bottlenecks and
optimizes resource utilization in distributed systems.

In summary, caching is essential in distributed systems for optimizing performance, ensuring

data consistency, and enhancing scalability by alleviating network congestion and reducing
response times.

Or
a) Comment on Concerns Regarding CPU Efficiency in Web-Based Distributed Computing:

It is understandable that the technical manager is concerned about the transition to web-based
distributed computing from systems that were originally designed to be CPU-efficient. Web-based
distributed computing involves communication between different components over a network, which
can introduce additional overhead and resource utilization compared to traditional CPU-bound
applications. The increased network latency, data serialization/deserialization, and coordination among
distributed components can potentially impact CPU efficiency.

However, the concerns are not unfounded. In a distributed system, the focus shifts from optimizing CPU
performance to optimizing network communication, data transfer, and overall system scalability. While
CPU efficiency remains important, other factors such as latency, bandwidth, and fault tolerance become
equally significant in distributed systems. As a software engineer, it is essential to strike a balance
between CPU efficiency and overall system performance to ensure the successful transition to web-
based distributed computing.

b) Synchronization Challenges in Heterogeneous Distributed Systems:

In a distributed system where components are very heterogeneous, synchronization becomes a more
significant problem due to the following reasons:

1. Different Architectures: Heterogeneous components may run on different platforms or have varying
hardware configurations, making it challenging to synchronize data and operations effectively.

2. Diverse Communication Protocols: Heterogeneous systems may use different communication

protocols or data formats, complicating the exchange of information and coordination between
components.

3. Varying Processing Speeds: Components with different processing speeds or capabilities may lead to
synchronization issues, such as race conditions or data inconsistency.

4. Time Synchronization: Ensuring accurate time synchronization across heterogeneous components is

crucial for coordinating events and maintaining system consistency.

To address synchronization challenges in heterogeneous distributed systems, software engineers must

design robust synchronization mechanisms, use standard communication protocols, implement data
serialization techniques, and ensure proper error handling and conflict resolution strategies.

c) Importance of Caching in Distributed Systems:

Caching plays a crucial role in improving performance, reducing latency, and avoiding inconsistent or
stale data in distributed systems by:

1. Reducing Network Traffic: Caching frequently accessed data or results locally at each node helps
reduce the need to fetch data over the network, thereby decreasing network traffic and latency.
2. Enhancing Scalability: Caching can help distribute the load evenly by serving cached data quickly
without overloading the backend servers, leading to better scalability and resource utilization.

3. Improving Response Time: Cached data can be served quickly to clients, improving response times
and overall system performance.

4. Mitigating Data Inconsistency: By maintaining a consistent cache invalidation strategy and updating
cached data regularly, caching helps avoid serving stale or outdated information to clients.

5. Enhancing Availability: Caching can improve the availability of data by serving cached copies even in
the event of network failures or server downtime, ensuring uninterrupted service for users.

Overall, caching in distributed systems is essential for optimizing performance, reducing resource
consumption, improving user experience, and maintaining data consistency across the system. Proper
caching strategies are vital for maximizing the benefits of distributed computing while minimizing
potential drawbacks related to data inconsistency or staleness.

Question 6

a) Describe a peer-to-peer architectural model for construction of a named distributed system of your
choice.|8|

b) Describe the asymmetric cryptography technique and show how it can be used in supporting
security in distributed systems.18|

c) Explain synchronization with respect to distributed systems

According to www.iAsk.ai Ask Ai Questions Search Engine:

Answer:
a) Peer-to-Peer Architectural Model for a Distributed System:

A peer-to-peer (P2P) architectural model for constructing a distributed system involves nodes
that act both as clients and servers, enabling decentralized communication and resource sharing
among participants. One example of a P2P distributed system is the implementation of a file-
sharing network using a structured overlay network with a Distributed Hash Table (DHT).

In this model:

1. Node Setup: Each node in the network functions as both a client and a server, capable of
initiating requests for resources and responding to requests from other nodes.
 2. Overlay Network: The P2P network establishes an overlay network on top of the
physical network topology, allowing direct communication between peers via logical
links.
3. Resource Discovery: Nodes use the DHT to efficiently search for resources by assigning
ownership of files to specific peers based on consistent hashing.
4. Routing: Nodes maintain lists of neighbors to facilitate efficient routing of traffic
through the network, ensuring that any node can search for resources effectively.

By implementing this P2P architectural model with a structured overlay network and DHT, the
distributed system can provide efficient resource discovery and sharing capabilities while
maintaining decentralization and robustness in the face of node churn.

b) Asymmetric Cryptography Technique for Security in Distributed Systems:

Asymmetric cryptography, also known as public-key cryptography, utilizes key pairs

consisting of a public key and a private key. The public key is shared openly, while the private
key is kept secret. This technique can support security in distributed systems through various
mechanisms:

1. Secure Communication: In a distributed system, nodes can use asymmetric

cryptography to establish secure communication channels. Messages encrypted with a
recipient’s public key can only be decrypted using their corresponding private key,
ensuring confidentiality.
2. Digital Signatures: Nodes can sign messages with their private keys to create digital
signatures. Recipients can verify the authenticity of the sender by decrypting the
signature with the sender’s public key, providing integrity and non-repudiation.
3. Key Exchange: Asymmetric cryptography enables secure key exchange protocols like
Diffie-Hellman, allowing nodes to negotiate session keys without exposing them to
potential eavesdroppers.
4. Authentication: Public-key certificates can be used to authenticate nodes in a distributed
system, ensuring that participants are who they claim to be.

By leveraging asymmetric cryptography techniques, distributed systems can enhance data

security, privacy, and trust among interconnected nodes.

c) Synchronization in Distributed Systems:

Synchronization in distributed systems refers to the coordination of activities across multiple

nodes to ensure consistency and coherence in shared data or processes. Key aspects of
synchronization include:

1. Concurrency Control: Distributed systems often involve parallel execution of tasks

across nodes. Synchronization mechanisms such as locks, semaphores, and transactions
help manage concurrent access to shared resources to prevent conflicts and maintain data
integrity.
 2. Consistency Models: Distributed systems may adopt different consistency models like
eventual consistency or strong consistency based on application requirements.
Synchronization protocols ensure that updates are propagated correctly among nodes to
maintain the desired level of consistency.
3. Clock Synchronization: Clock synchronization is crucial in distributed systems to order
events accurately across nodes. Algorithms like NTP (Network Time Protocol)
synchronize clocks to establish a common time reference for coordination.
4. Data Replication: Synchronization mechanisms govern data replication strategies to
ensure that replicas across nodes remain consistent and up-to-date, minimizing
divergence or conflicts.

Effective synchronization mechanisms are essential for ensuring reliability, fault tolerance, and
performance optimization in distributed systems by managing interactions among decentralized
components.

Or

a) Peer-to-Peer Architectural Model for a Distributed System:

A peer-to-peer (P2P) architectural model for constructing a distributed system involves
nodes that act both as clients and servers, enabling decentralized communication and
resource sharing among participants. One example of a P2P distributed system is the
implementation of a file-sharing network using a structured overlay network with a
Distributed Hash Table (DHT).

In this model:

1. Node Setup: Each node in the network functions as both a client and a server, capable
of initiating requests for resources and responding to requests from other nodes.
2. Overlay Network: The P2P network establishes an overlay network on top of the
physical network topology, allowing direct communication between peers via logical
links.
3. Resource Discovery: Nodes use the DHT to efficiently search for resources by
assigning ownership of files to specific peers based on consistent hashing.
4. Routing: Nodes maintain lists of neighbors to facilitate efficient routing of traffic through
the network, ensuring that any node can search for resources effectively.

By implementing this P2P architectural model with a structured overlay network and
DHT, the distributed system can provide efficient resource discovery and sharing
capabilities while maintaining decentralization and robustness in the face of node churn.

b) Asymmetric Cryptography Technique for Security in

Distributed Systems:
 Asymmetric cryptography, also known as public-key cryptography, utilizes key pairs
consisting of a public key and a private key. The public key is shared openly, while the
private key is kept secret. This technique can support security in distributed systems
through various mechanisms:

1. Secure Communication: In a distributed system, nodes can use asymmetric

cryptography to establish secure communication channels. Messages encrypted with a
recipient’s public key can only be decrypted using their corresponding private key,
ensuring confidentiality.
2. Digital Signatures: Nodes can sign messages with their private keys to create digital
signatures. Recipients can verify the authenticity of the sender by decrypting the
signature with the sender’s public key, providing integrity and non-repudiation.
3. Key Exchange: Asymmetric cryptography enables secure key exchange protocols like
Diffie-Hellman, allowing nodes to negotiate session keys without exposing them to
potential eavesdroppers.
4. Authentication: Public-key certificates can be used to authenticate nodes in a
distributed system, ensuring that participants are who they claim to be.

By leveraging asymmetric cryptography techniques, distributed systems can enhance

data security, privacy, and trust among interconnected nodes.

c) Synchronization in Distributed Systems:

Synchronization in distributed systems refers to the coordination of activities across
multiple nodes to ensure consistency and coherence in shared data or processes. Key
aspects of synchronization include:

1. Concurrency Control: Distributed systems often involve parallel execution of tasks

across nodes. Synchronization mechanisms such as locks, semaphores, and
transactions help manage concurrent access to shared resources to prevent conflicts
and maintain data integrity.
2. Consistency Models: Distributed systems may adopt different consistency models like
eventual consistency or strong consistency based on application requirements.
Synchronization protocols ensure that updates are propagated correctly among nodes
to maintain the desired level of consistency.
3. Clock Synchronization: Clock synchronization is crucial in distributed systems to order
events accurately across nodes. Algorithms like NTP (Network Time Protocol)
synchronize clocks to establish a common time reference for coordination.
4. Data Replication: Synchronization mechanisms govern data replication strategies to
ensure that replicas across nodes remain consistent and up-to-date, minimizing
divergence or conflicts.

Effective synchronization mechanisms are essential for ensuring reliability, fault

tolerance, and performance optimization in distributed systems by managing
interactions among decentralized components.
Question 1

Cloud computing services like Amazon's EC2 assign users virtual machines (VMs) instead of allocating
physical machines directly. Doing so provides at least three major benefits to Amazon.

a) Discuss these three benefits, giving a brief motivation for each one.

b) Explain the benefits of using cloud computing for a supermarket chain.

c) Explain three differences between grid computing and cluster computing

a) Three Benefits of Using Virtual Machines in Cloud Computing:

1. Cost Efficiency: By assigning virtual machines (VMs) instead of physical machines, cloud computing
services like Amazon's EC2 can optimize resource utilization and reduce operational costs. VMs allow for
efficient allocation of computing resources based on demand, enabling users to pay for only the
resources they use. This pay-as-you-go model helps businesses save money by avoiding upfront
infrastructure investments and scaling resources as needed, leading to cost efficiency and scalability.

2. Flexibility and Scalability: Virtual machines offer flexibility in terms of resource allocation,
configuration, and deployment. Users can easily create, modify, and resize VM instances based on their
requirements without the limitations of physical hardware. Cloud providers like Amazon can quickly
provision and scale VMs to meet varying workloads, ensuring high availability and performance. This
flexibility and scalability enable businesses to adapt to changing demands and optimize resource
utilization efficiently.

3. Resource Isolation and Management: Virtual machines provide isolation between different users and
applications running on the same physical infrastructure, enhancing security and performance.
Amazon's EC2 can allocate VMs to users with dedicated computing resources and customized
configurations, ensuring the isolation of workloads and data. VM management tools enable efficient
monitoring, provisioning, and orchestration of VM instances, simplifying resource management and
optimizing workload distribution in a cloud environment.

b) Benefits of Using Cloud Computing for a Supermarket Chain:

1. Scalability: Cloud computing allows a supermarket chain to scale its IT infrastructure according to
seasonal demand, promotions, or new store openings. The supermarket chain can easily provision
additional resources, such as storage and computing power, to handle increased website traffic,
inventory management, customer data analysis, and online ordering services during peak times.
2. Cost Efficiency: Utilizing cloud services eliminates the need for the supermarket chain to invest in and
maintain on-premises hardware and infrastructure. Cloud computing offers a pay-as-you-go model,
enabling the supermarket chain to pay for only the resources they consume, reducing operational costs,
and avoiding unnecessary capital expenditures on IT infrastructure.

3. Data Security and Backup: Cloud computing provides robust data security measures, such as
encryption, access control, and data backup, ensuring the protection and integrity of sensitive customer
information and business data. The supermarket chain can leverage cloud-based data storage and
backup services to prevent data loss, mitigate risks of hardware failures, and enhance disaster recovery
capabilities.

c) Differences Between Grid Computing and Cluster Computing:

1. Architecture: Grid computing involves coordinating resources from multiple distributed and
independent sources to solve large-scale computational problems, focusing on resource sharing and
collaboration. In contrast, cluster computing connects multiple computers or servers within a local
network to work on a single task, emphasizing parallel processing and high-performance computing
within a centralized system.

2. Resource Ownership: In grid computing, resources are owned by different organizations or

individuals, and cooperation is based on a shared infrastructure for resource utilization. Cluster
computing typically involves a single organization or entity owning and managing the cluster of
machines, leading to centralized control and resource allocation within the cluster.

3. Task Allocation: Grid computing assigns tasks to available resources dynamically based on resource
availability and capabilities across the grid network, enabling efficient load balancing and optimization.
Cluster computing allocates tasks to specific nodes within the cluster based on predefined
configurations, ensuring dedicated resources and performance isolation for each task or application
running on the cluster.

Question 2

a) Discuss the design of a distributed system that implements the principles of Edge Computing.

b) Discuss how recent developments in hardware and communication technologies have contributed
to the emergence Internet of Things (loT)
a) Design of a Distributed System Implementing Edge Computing Principles:

Edge computing involves moving computation and data processing closer to the edge of the network,
near where data is generated, rather than relying solely on centralized cloud servers. A distributed
system designed with edge computing principles would involve deploying computing resources (such as
servers, storage, and processing units) at the network edge to enable faster data processing, reduce
latency, and enhance overall system efficiency. The design of such a distributed system could include the
following components:

1. Edge Devices: Devices such as sensors, cameras, and IoT devices that generate data at the network
edge.

2. Edge Servers: Servers deployed at the edge of the network to process data locally and handle
computing tasks closer to where data is generated.

3. Edge Gateways: Devices that connect edge devices to the edge servers and facilitate data transfer and
communication between devices and servers.

4. Distributed Data Storage: Storage systems distributed at the edge to store and manage data closer to
where it is generated, reducing latency and improving data access.

5. Edge Computing Software: Software applications and services designed to run on edge servers to
process data, run analytics, and enable real-time decision-making at the network edge.

6. Communication Protocols: Protocols and standards for efficient communication between edge
devices, gateways, servers, and cloud resources, ensuring seamless data flow and system integration.

The design of a distributed system implementing edge computing principles aims to enhance
performance, reduce bandwidth usage, improve scalability, and support real-time applications by
leveraging distributed resources at the network edge.

b) Recent Developments in Hardware and Communication Technologies Contributing to IoT:

1. Miniaturization and Low Power Consumption: Recent advancements in hardware technology have led
to the development of small, low-power devices with enhanced computing capabilities. Miniaturized
sensors, processors, and communication modules enable the creation of IoT devices that can operate
efficiently on battery power for extended periods, supporting the proliferation of IoT applications in
various industries.

2. Wireless Communication Standards: The evolution of wireless communication technologies, such as

Bluetooth Low Energy (BLE), Zigbee, LoRa, and 5G networks, has enabled seamless connectivity and data
exchange between IoT devices and networks. These communication standards support reliable data
transmission, low latency, and improved network coverage, facilitating the growth of IoT ecosystems
and enabling diverse IoT solutions.
3. Edge Computing Infrastructure: The rise of edge computing architectures and edge devices with
enhanced processing capabilities has contributed to the development of IoT systems that can perform
data processing and analytics locally, reducing latency and enhancing real-time decision-making. Edge
computing infrastructure complements cloud services by offloading computation tasks to edge servers,
supporting IoT applications that require low latency, high availability, and efficient resource utilization.

4. Security and Privacy Measures: Recent advancements in hardware encryption, secure boot
mechanisms, and firmware updates have strengthened the security and privacy features of IoT devices
and networks. Hardware-based security solutions, such as Trusted Platform Modules (TPMs) and Secure
Element chips, protect IoT devices from cyber threats, unauthorized access, and data breaches, ensuring
the integrity and confidentiality of IoT data and communications.

Overall, recent developments in hardware and communication technologies have accelerated the
growth of IoT ecosystems, driving innovation, improving connectivity, and enabling diverse IoT
applications across industries such as healthcare, smart homes, transportation, and industrial
automation. Or

According to www.iAsk.ai Ask Ai Questions Search Engine:

Design of a Distributed System Implementing Edge Computing Principles:

To design a distributed system that implements the principles of Edge Computing, several key
considerations need to be taken into account:

1. Location of Compute and Storage Resources: The core principle of edge computing is
to place computing and storage resources closer to the data source. In the design, identify
the specific locations where data is generated and ensure that compute and storage
resources are deployed in proximity to these sources.
2. Network Architecture: Develop a network architecture that enables seamless
communication between edge devices and central data centers. This architecture should
support low-latency data transmission, efficient data processing at the edge, and secure
data transfer back to the central data center.
3. Scalability: Design the distributed system to be scalable, allowing for the easy addition
of new edge devices as the network grows. Scalability ensures that the system can handle
increasing amounts of data and computational tasks without compromising performance.
4. Data Processing and Analysis: Implement algorithms and software tools for real-time
data processing and analysis at the edge. This includes predictive analytics, machine
learning models, and other tools that can derive actionable insights from the data
collected at the edge.
 5. Security Measures: Incorporate robust security measures to protect data at both the edge
and central data centers. Encryption, access controls, authentication mechanisms, and
secure communication protocols are essential components of a secure distributed system.
6. Fault Tolerance: Ensure that the distributed system is fault-tolerant, meaning it can
continue operating even if individual edge devices fail. Redundancy, failover
mechanisms, and automated recovery processes should be built into the design to
maintain system reliability.
7. Interoperability: Address interoperability challenges by standardizing communication
protocols, data formats, and interfaces across different edge devices. Compatibility
between various components of the distributed system is crucial for seamless operation.

By incorporating these design elements into a distributed system architecture, organizations can
effectively leverage edge computing principles to enhance data processing efficiency, reduce
latency, improve scalability, and enable real-time decision-making in diverse applications.

Recent Developments in Hardware and Communication Technologies Contributing to IoT:

Recent advancements in hardware and communication technologies have significantly

contributed to the emergence and growth of the Internet of Things (IoT). Some key
developments include:

1. Powerful Edge Devices: The proliferation of powerful edge computing devices with
enhanced processing capabilities has enabled more complex computations to be
performed at the edge of networks. These devices can process large volumes of data
locally, reducing reliance on centralized cloud infrastructure.
2. 5G Networks: The rollout of 5G networks has revolutionized connectivity by providing
ultra-fast speeds, low latency, and high bandwidth capacity. 5G technology facilitates
real-time communication between IoT devices, enabling rapid data transfer and
enhancing IoT applications such as autonomous vehicles, smart cities, and industrial
automation.
3. AI Integration: The integration of Artificial Intelligence (AI) and Machine Learning
(ML) algorithms into IoT systems has unlocked new possibilities for intelligent decision-
making and predictive analytics. AI-powered IoT devices can autonomously analyze data
streams, detect patterns, optimize processes, and improve overall system efficiency.
4. Edge Computing Infrastructure: The development of robust edge computing
infrastructure has facilitated decentralized data processing in IoT ecosystems. By
distributing computational tasks across edge devices, organizations can achieve faster
response times, reduced network congestion, improved reliability, and enhanced security
in IoT deployments.

These recent developments in hardware and communication technologies have played a pivotal
role in advancing the capabilities of IoT systems, enabling innovative applications across
industries while addressing challenges related to scalability, latency, security, and real-time data
processing.
Question 3

a) Describe the properties of the Distributed Ledger Technology (DLT) [15]

b) Discuss the design of a system implementing the Distributed Ledger Technology.Your discussion
should highlight the advantages of DLT over ordinary distributed systems.[10]

According to www.iAsk.ai Ask Ai Questions Search Engine:

Properties of Distributed Ledger Technology (DLT)

Distributed Ledger Technology (DLT) possesses several key properties that distinguish it from
traditional centralized systems. These properties include:

1. Decentralization: DLT operates in a decentralized manner where multiple nodes

maintain the ledger independently, ensuring no single point of control or failure.
2. Immutability: Data stored on a DLT is cryptographically secured, making it tamper-
proof and immutable. Once information is recorded, it cannot be altered or deleted.
3. Append-Only: DLTs are append-only databases, meaning that new data can be added
but existing data cannot be modified or removed.
4. Distributed Consensus: Changes to the ledger are validated through a consensus
mechanism involving all participating nodes, ensuring agreement on the accuracy of
transactions.
5. Transparency: All participants in a DLT network have visibility into the transactions
and changes occurring on the ledger, promoting trust and accountability.
6. Smart Contracts: DLT supports smart contracts, self-executing agreements coded into
the system, enabling automated and secure transactions based on predefined conditions.
7. Fault Tolerance: Due to its distributed nature, DLT systems are resilient to failures or
attacks on individual nodes, ensuring data integrity and availability.
8. Shared Ownership: No single entity controls the entire ledger; instead, ownership is
distributed among network participants, fostering a collaborative environment.
9. Efficiency: DLT enables fast and efficient transaction processing by eliminating
intermediaries and streamlining verification processes across the network.
10. Security: Utilizing cryptographic techniques, DLT ensures high levels of security for
data storage and transaction validation, reducing the risk of fraud or unauthorized access.

Design of a System Implementing Distributed Ledger Technology

Implementing a system based on Distributed Ledger Technology involves designing a network

where multiple nodes collaborate to maintain a shared ledger with the following advantages over
ordinary distributed systems:

1. Enhanced Security: DLT’s cryptographic mechanisms provide robust security measures

that protect data integrity and prevent unauthorized modifications, offering higher
security compared to traditional distributed systems.
 2. Improved Transparency: The transparent nature of DLT ensures that all participants
have real-time visibility into transactions and changes made to the ledger, enhancing trust
and accountability within the system.
3. Decentralized Governance: By distributing control among network nodes, DLT systems
eliminate single points of failure or control, promoting democratic decision-making
processes and reducing vulnerabilities.
4. Efficient Transaction Processing: DLT’s streamlined consensus mechanisms enable
faster transaction processing times by eliminating intermediaries and automating
verification processes across the network.
5. Smart Contract Capabilities: Leveraging smart contracts within a DLT system allows
for automated execution of predefined agreements based on specific conditions, reducing
manual intervention and enhancing operational efficiency.
6. Resilience to Failures: The fault-tolerant nature of DLT ensures that even if individual
nodes fail or are compromised, the system remains operational and data remains
accessible through other network participants.
7. Immutable Data Storage: By storing data in an immutable manner using cryptographic
techniques, DLT systems prevent data tampering or unauthorized alterations, maintaining
data integrity over time.
8. Scalability and Flexibility: DLT systems can scale efficiently to accommodate growing
transaction volumes while remaining adaptable to changing business requirements
without compromising performance or security.
9. Cost-Effective Operations: By reducing the need for intermediaries and automating
processes through smart contracts, DLT systems offer cost-effective solutions for
transaction processing and data management.
10. Regulatory Compliance: DLT systems can be designed to incorporate regulatory
requirements into their operations, ensuring compliance with industry standards and legal
frameworks while maintaining transparency and auditability.

In conclusion, implementing Distributed Ledger Technology offers numerous advantages over

traditional distributed systems by enhancing security, transparency, efficiency, governance,
resilience, and cost-effectiveness in managing transactions and data within a decentralized
network environment.

Question 4

a) Given a bag of words in Figure 4, discuss how MapReduce works.[

Cloud computing is not only the future of computing but the present and the entire past of
computing]

Figure 4

I'm unable to view Figure 4 as it was not provided in the question. However, I can explain how
MapReduce works in the context of processing a bag of words.
MapReduce is a programming model and processing framework used to process large datasets in a
distributed computing environment. The basic idea behind MapReduce is to break down a complex
computation task into smaller, independent tasks that can be processed in parallel across multiple
nodes in a cluster. The MapReduce framework consists of two main phases: the Map phase and the
Reduce phase.

1. Map Phase:

- In the Map phase, the input data (in this case, the bag of words) is divided into key-value pairs. Each
key-value pair represents a unit of data that will be processed independently.

- The Map function processes each key-value pair and generates intermediate key-value pairs as output.
For a bag of words, the Map function can tokenize the input text into individual words and assign a
count of 1 to each word as the value.

- The intermediate key-value pairs are then shuffled and sorted based on their keys to group together
values associated with the same key.

2. Shuffle and Sort:

- The shuffle and sort phase reorganizes the intermediate key-value pairs generated by the Map
function. All values with the same key from different nodes are grouped together.

- This phase ensures that all values corresponding to the same key are processed together during the
Reduce phase.

3. Reduce Phase:

- In the Reduce phase, each unique key and its corresponding list of values are passed to the Reduce
function. The Reduce function aggregates and processes these values to produce the final output.

- For a bag of words, the Reduce function can sum up the word counts for each unique word to generate
a final word frequency count.

By utilizing the MapReduce framework, the processing of a bag of words dataset can be distributed
across multiple nodes in a cluster, allowing for parallel computation and efficient processing of large
volumes of data. MapReduce simplifies the task of processing and analyzing big data by breaking it down
into smaller, manageable chunks that can be processed in parallel, leading to faster computation and
scalability.
b) Distinguish between Hadoop 1.0 and Hadoop 2.0.

c) Discuss why replication is pursued in the HDFS though it may cause data redundancy.

According to www.iAsk.ai Ask Ai Questions Search Engine:

Distinguish between Hadoop 1.0 and Hadoop 2.0:

Hadoop 1.0 and Hadoop 2.0 are two significant versions of the Hadoop framework, each with
distinct features and improvements:

1. Components:
o Hadoop 1.0 primarily includes MapReduce for data processing, while Hadoop 2.0
introduces YARN (Yet Another Resource Negotiator) along with MapReduce
version 2. YARN enhances resource management capabilities and allows for
more diverse workloads.
2. Daemons:
o In Hadoop 1.0, the architecture follows a Master-Slave model with a single master
node and multiple slave nodes. On the other hand, Hadoop 2.0 features multiple
masters (active namenodes and standby namenodes) along with multiple slaves,
providing better fault tolerance.
3. Working:
o Hadoop 1.0 relies on HDFS for storage and MapReduce for resource management
and data processing, leading to performance limitations due to the workload on
MapReduce. In contrast, Hadoop 2.0 maintains HDFS for storage and utilizes
YARN for resource management, which allocates resources efficiently without
overburdening MapReduce.
4. Limitations:
o Hadoop 1.0 suffers from a single point of failure issue with its Master-Slave
architecture, where the failure of the master node can disrupt the entire cluster.
Hadoop 2.0 addresses this by incorporating multiple masters and standby nodes,
ensuring high availability and eliminating single points of failure.
5. Ecosystem:
o Both versions support various tools like Oozie for workflow scheduling, Pig,
Hive, Mahout for data processing, Sqoop for structured data import/export, and
Flume for unstructured data handling.
6. Windows Support:
o While Hadoop 1.0 lacks official support for Microsoft Windows by Apache,
Hadoop 2.0 offers compatibility with Windows environments.

Discuss why replication is pursued in the HDFS though it may cause data redundancy:

Replication in the Hadoop Distributed File System (HDFS) serves a crucial purpose despite
potentially leading to data redundancy:

 Fault Tolerance: Replication ensures high availability and fault tolerance in distributed
systems like HDFS by storing data copies across multiple nodes. If one replica becomes
 unavailable or corrupted, the system can retrieve the data from other replicas, preventing
data loss.
 Reliability: By replicating data across different nodes, HDFS reduces the risk of losing
information due to hardware failures or network issues. This redundancy enhances the
reliability of data storage and retrieval operations.
 Data Accessibility: Replication facilitates faster access to data by allowing parallel reads
from multiple replicas simultaneously. This improves overall system performance and
responsiveness when handling large datasets.
 Load Balancing: Replication also aids in load distribution across nodes in the cluster,
preventing hotspots and optimizing resource utilization within the system.

While replication does introduce some level of data redundancy and increased storage
requirements, these trade-offs are justified by the benefits it provides in terms of system
reliability, fault tolerance, and performance optimization.

Question 5

3) The construction of distributed systems produces many challenges. Discuss any of the five main
challenges and how they can be addressed.

b) Discuss the typical design goals of distributed systems.

According to www.iAsk.ai Ask Ai Questions Search Engine:

Challenges in Distributed Systems:

Concurrency: Concurrency is a significant challenge in distributed systems as multiple clients

may attempt to access shared resources simultaneously, leading to potential conflicts and data
inconsistency. To address this challenge, techniques such as locking mechanisms, transactions,
and version control can be implemented. Locking mechanisms ensure that only one client can
access a resource at a time, preventing concurrent modifications. Transactions provide a way to
group multiple operations into an atomic unit, ensuring either all operations succeed or none are
applied. Version control helps track changes to shared data and resolve conflicts that may arise
due to concurrent access.

Security: Security is another critical challenge in distributed systems due to the need to protect
sensitive data and resources from unauthorized access and malicious attacks. To address security
concerns, encryption techniques can be employed to secure data transmission and storage.
Access control mechanisms, such as authentication and authorization, help ensure that only
authorized users can access specific resources. Additionally, implementing firewalls, intrusion
detection systems, and regular security audits can enhance the overall security posture of a
distributed system.
Failure Handling: Handling failures in distributed systems poses a significant challenge as
components may fail independently, leading to partial system failures or incorrect results. To
address this challenge, fault tolerance mechanisms like redundancy, replication, and
checkpointing can be implemented. Redundancy involves duplicating critical components to
ensure system availability even if some components fail. Replication allows data to be stored on
multiple nodes to prevent data loss in case of node failure. Checkpointing involves saving the
state of the system periodically so that it can be restored in case of failure.

Scalability: Scalability is a key design goal for distributed systems to ensure that the system can
handle increasing loads and users without sacrificing performance. To achieve scalability,
techniques such as load balancing, horizontal scaling, and partitioning can be employed. Load
balancing distributes incoming requests evenly across multiple servers to prevent overloading
any single server. Horizontal scaling involves adding more nodes to the system to increase
capacity. Partitioning allows data to be divided into smaller subsets distributed across nodes for
efficient processing.

Typical Design Goals of Distributed Systems:

Heterogeneity: One of the typical design goals of distributed systems is handling heterogeneity
effectively. This involves supporting diverse hardware platforms, operating systems, and
network configurations seamlessly. Middleware plays a crucial role in addressing heterogeneity
by providing a common interface for applications to interact with different components in the
distributed system.

Openness: Another design goal is openness, which focuses on enabling easy integration of new
services and resources into the distributed system. Open systems have well-defined interfaces
that allow for interoperability between different components regardless of their underlying
technologies.

Scalability: Scalability is a fundamental design goal aimed at ensuring that the system can grow
in terms of users, resources, and workload without compromising performance or reliability.
Designing for scalability involves considering factors like load distribution, resource allocation,
and fault tolerance mechanisms.

Security: Ensuring security is a critical design goal for distributed systems to protect sensitive
data from unauthorized access and maintain the integrity of the system. Security measures such
as encryption, authentication, authorization, and secure communication protocols are essential
components of designing secure distributed systems.

Failure Handling: Effective failure handling is another key design goal that focuses on building
resilience against component failures and ensuring continuity of service even in the presence of
faults. Techniques like fault tolerance, redundancy, error recovery mechanisms, and graceful
degradation are essential for robust failure handling in distributed systems.

In summary, addressing challenges such as concurrency, security, failure handling, and

scalability requires implementing appropriate mechanisms like locking, encryption, fault
tolerance, redundancy, load balancing, and partitioning. Meanwhile, achieving design goals like
handling heterogeneity effectively, ensuring openness for easy integration of services/resources,
scalability for growth without performance degradation or reliability issues, maintaining security
through encryption/authentication/authorization measures while also focusing on effective failure
handling through fault tolerance/redundancy/error recovery mechanisms are crucial aspects when
designing distributed systems.

Question 6

•) Explain three benefits of middleware to distributed systems programmers, providing an example

for each benefit

b) Desiribe a pocr-to-poet system contraiting it with a chent-sen er system.

* Describe the gossip protocol.

According to www.iAsk.ai Ask Ai Questions Search Engine:

Three Benefits of Middleware to Distributed Systems Programmers:

1. Enhanced Scalability: Middleware provides distributed systems programmers with the

ability to scale their applications efficiently. An example of this benefit is seen in the use
of message-oriented middleware (MOM) in a distributed system handling a large volume
of transactions. MOM allows for asynchronous communication between components,
enabling the system to handle varying loads without impacting performance. For
instance, in an e-commerce platform, MOM can manage order processing and inventory
updates asynchronously, ensuring smooth scalability during peak shopping seasons.
2. Improved Interoperability: Middleware facilitates seamless communication and
interaction between heterogeneous systems and applications. For instance, an enterprise
service bus (ESB) as a middleware solution can integrate various backend systems such
as customer relationship management (CRM) and enterprise resource planning (ERP)
systems. This interoperability enables data exchange and process orchestration across
different platforms, enhancing overall system efficiency and functionality.
3. Enhanced Security: Middleware offers security features that help protect distributed
systems from unauthorized access and data breaches. For example, middleware
components like security tokens or encryption mechanisms can secure communication
channels between different system nodes. In a healthcare application handling sensitive
patient data, middleware can enforce access control policies and encryption to ensure
data confidentiality and integrity across the distributed environment.

Contrasting Point-to-Point System with Client-Server System:

In a point-to-point system, communication occurs directly between two endpoints without

intermediary components. This setup is simple and efficient for direct interactions but lacks
scalability and flexibility when adding more endpoints or functionalities. An example is a peer-
to-peer file sharing network where users directly exchange files without central servers.
On the other hand, a client-server system involves clients requesting services or resources from
centralized servers. This architecture allows for better management, scalability, and security
compared to point-to-point systems. For instance, in a web application, clients (browsers)
interact with a central server hosting the application logic and data storage, enabling efficient
resource sharing and access control.

Gossip Protocol:

The gossip protocol is a communication algorithm used in distributed systems for disseminating
information among nodes in a decentralized manner. In this protocol, nodes randomly select
peers to exchange information with, spreading updates or data throughout the network. Each
node shares received information with other nodes it interacts with, leading to rapid
dissemination of updates across the system.

The gossip protocol is commonly employed in scenarios where eventual consistency is

acceptable, such as in peer-to-peer networks or large-scale distributed databases. It helps
maintain synchronization and coherence among distributed nodes by propagating changes
through indirect interactions rather than relying on centralized coordination.

Question 1

a) Using an application you are familiar with, define a distributed syston.

b) Describe how distributed systems are able to project a "single system image".

e) In a distributed system, "Callures occur and we may not know or be fold". Discuss how this is so.

1) Describe how fault masking is achieved in distributed systems.

a) Using an application you are familiar with, define a distributed system:

A common example of a distributed system that many people are familiar with is a cloud
storage service like Google Drive. In this context, a distributed system refers to the
infrastructure where data is stored and accessed across multiple servers or data
centers. Google Drive allows users to store files and access them from various devices
connected to the internet. The system ensures data availability and reliability by
distributing copies of files across different servers, enabling users to access their data
seamlessly from anywhere.

b) Describe how distributed systems are able to project a “single system image”:

Distributed systems achieve the concept of a “single system image” by presenting a

unified view to users despite the underlying complexity of multiple interconnected
components. This illusion of a single cohesive system is maintained through various
mechanisms such as:
 Transparency: Distributed systems use transparency mechanisms to hide the
distribution of resources and present them as a unified entity. This includes location
transparency (users are unaware of the physical location of resources), access
transparency (users access resources uniformly), and failure transparency (users are
shielded from failures).

 Resource Management: Middleware plays a crucial role in managing resources across

distributed systems. It abstracts the underlying hardware and software differences,
providing a consistent interface for applications to interact with.

 Communication Protocols: Distributed systems rely on standardized communication

protocols to ensure seamless interaction between components. These protocols enable
different nodes to exchange information and coordinate their activities effectively.

By leveraging these mechanisms, distributed systems create a cohesive user

experience where users interact with the system as if it were a single entity, despite the
underlying complexity of multiple interconnected components.

c) In a distributed system, failures occur and we may not know or be told.

Discuss how this is so:

In distributed systems, failures are inevitable due to the complex nature of

interconnected components spread across multiple nodes. Several factors contribute to
the challenge of detecting failures in distributed systems:

 Asynchronous Communication: Distributed systems often operate asynchronously,

meaning that components do not share a global clock or synchronize their actions. This
lack of synchronization makes it challenging to detect failures promptly since there is no
central point coordinating all activities.

 Partial Failures: Failures in distributed systems can be partial, affecting only specific
components or nodes without causing a complete system outage. These partial failures
may not be immediately apparent to users or other components, leading to undetected
issues.

 Network Partitioning: Network partitions can isolate certain nodes or segments of the
distributed system, making it difficult for other components to detect whether a failure
has occurred in the isolated segment. This can result in silent failures that go unnoticed
until they impact system functionality.

Due to these complexities inherent in distributed systems, failures can occur without
immediate detection or notification, posing challenges for system administrators in
identifying and addressing issues proactively.

d) Describe how fault masking is achieved in distributed systems:

 Fault masking in distributed systems refers to the ability of the system to isolate and
contain faults without impacting the overall functionality or availability of the entire
system. Several techniques are employed to achieve fault masking:

 Redundancy: Distributed systems often incorporate redundancy at various levels, such

as data replication, server mirroring, or task duplication. By maintaining redundant
copies of data or services, the system can continue functioning even if one component
fails.

 Failover Mechanisms: Distributed systems implement failover mechanisms that

automatically redirect traffic or workload from a failed component to an operational one.
This seamless transition helps mask faults by ensuring uninterrupted service provision.

 Load Balancing: Load balancing distributes incoming network traffic evenly across
multiple servers or resources in a distributed system. By spreading out the workload,
load balancing helps prevent individual components from becoming overwhelmed and
failing under excessive demand.

 Monitoring and Recovery: Continuous monitoring tools track the health and
performance of components in real-time. When anomalies or failures are detected,
automated recovery processes kick in to restore functionality and mask faults from end-
users.

Question 2

a) Discuss the role of naming services in distributed systems.

b) Discuss the operation of a typical domain name server (DNS) using a saltable example

c) Describe two navigation schemes that can be used for name resolution in domain name systems.

Role of Naming Services in Distributed Systems:

Naming services play a crucial role in distributed systems by providing a way to uniquely
identify resources within the network. They help in abstracting the physical locations and details
of these resources, making it easier for users and applications to access them without needing to
know specific network addresses. In a distributed system, naming services enable the following
key functionalities:
 1. Resource Identification: Naming services allow resources such as files, devices, or
services to be given meaningful names that can be used to locate and access them across
the network.
2. Location Transparency: By using names instead of direct addresses, naming services
provide location transparency, allowing resources to be moved or replicated without
affecting how they are accessed.
3. Scalability: Naming services help in scaling distributed systems by providing a
structured way to manage and organize a large number of resources efficiently.
4. Fault Tolerance: Distributed naming services can incorporate fault-tolerant mechanisms
to ensure continued availability even in the face of network failures or resource
unavailability.
5. Security: Naming services can also play a role in enforcing security policies by
controlling access to resources based on their names and associated permissions.

Overall, naming services act as a fundamental component in distributed systems, enabling

seamless resource access and management while abstracting the complexities of network
communication.

Operation of a Typical Domain Name Server (DNS) with an Example:

A typical Domain Name Server (DNS) operates by translating domain names into IP addresses
through a hierarchical system of servers. Here is an example illustrating the operation of a DNS
server:

1. User Action: A user enters a domain name like “example.com” into their web browser.
2. Recursive DNS Query: The browser sends a recursive DNS query to the ISP’s recursive
resolver to find the corresponding IP address for “example.com.”
3. Server Resolution: If the recursive resolver does not have the IP address cached, it
queries the root name servers, then TLD servers, and finally authoritative name servers
for “example.com” until it obtains the IP address.
4. Response: The authoritative name server for “example.com” provides the IP address
back to the recursive resolver, which then returns it to the user’s browser.
5. Caching: The recursive resolver caches the IP address for future requests, improving
efficiency for subsequent queries involving “example.com.”

This process showcases how DNS servers work collaboratively to translate domain names into
IP addresses, enabling users to access websites and other online resources seamlessly.

Navigation Schemes for Name Resolution in Domain Name Systems:

In domain name systems, there are two primary navigation schemes used for name resolution:

1. Iterative Resolution: In iterative resolution, the DNS server provides referrals to other
servers closer to the requested domain until it reaches an authoritative server that can
provide the final answer. This method involves multiple queries between servers but
allows for more control over the resolution process.
 2. Recursive Resolution: Recursive resolution involves the DNS server handling all
aspects of resolving a query on behalf of the client. The server itself queries other servers
as needed until it obtains the final answer and returns it directly to the client. This method
simplifies the process for clients but places more load on DNS servers.

These navigation schemes offer different approaches to resolving domain names into IP
addresses within DNS systems, catering to varying requirements based on efficiency and control
needs.

Question 3

a) Imagine you are a software cagineer in a company that is planning a move towards using web-
based distributed computing. The technical manager has called for a discussice of the challenpes
involved. The fact that the existing systems were originally designed with a vicw to being CPU-efficient
causes her particular concern.

She asks you to comment on whether her concerns are well-founded. Discuss whether you agree with
your technical manager.

b) Explain why peer-to-peer architectures (P2P) are said to lead naturally to balanced

(10)

loads and graceful scaling

b) Consider the Two Generals problem. Defend the assertion that it is impossible to design an
algorithm that is guaranteed to lead to a cocedinated solution.

According to www.iAsk.ai Ask Ai Questions Search Engine:

b) Peer-to-Peer Architectures (P2P) and Balanced Loads:

Peer-to-peer (P2P) architectures are known to naturally lead to balanced loads and graceful
scaling due to their decentralized nature. In a P2P network, each node can act as both a client and
a server, sharing resources directly with other nodes without the need for a central server. This
distributed approach helps in distributing the workload evenly across the network, leading to
balanced loads.

One of the key advantages of P2P architectures is that they are self-organizing and self-scaling.
As more nodes join the network, the overall capacity and resources of the system increase
organically without relying on a single point of failure. Additionally, P2P networks can
dynamically adapt to changes in the network by redistributing tasks and resources among nodes,
ensuring that the system remains balanced even as it scales.
The decentralized nature of P2P architectures also contributes to graceful scaling. Since there is
no central authority controlling the network, adding new nodes does not introduce bottlenecks or
single points of failure. Each node in a P2P network contributes to the overall processing power
and storage capacity, allowing the system to scale seamlessly without overloading any specific
node.

Overall, peer-to-peer architectures promote load balancing and graceful scaling by leveraging the
collective resources of all nodes in the network efficiently.

c) The Two Generals Problem and Coordinated Solution:

The Two Generals problem is a classic computer science issue that highlights the challenges of
achieving perfect coordination between two entities communicating over an unreliable channel.
In this scenario, two generals are planning a coordinated attack but must agree on the timing of
the assault without direct synchronous communication.

The assertion that it is impossible to design an algorithm that guarantees a coordinated solution
in the Two Generals problem stems from fundamental limitations in asynchronous
communication and uncertainty about message delivery in distributed systems. Due to factors
such as network delays, message loss, and lack of shared memory between the generals, there is
always a possibility of one general not receiving confirmation of the agreed-upon timing or
receiving conflicting information.

In essence, any solution proposed for the Two Generals problem must account for these
uncertainties and cannot provide an absolute guarantee of success due to the inherent challenges
posed by asynchronous communication across distributed systems. While certain probabilistic
algorithms or heuristics can improve the likelihood of reaching a coordinated solution, there will
always be a non-zero probability of failure or miscommunication in such scenarios.

Therefore, defending the assertion that designing an algorithm guaranteed to lead to a

coordinated solution in the Two Generals problem is indeed challenging due to the inherent
limitations of asynchronous communication and uncertainty in distributed environments.

Question 4

a) Compare and contrast loud computing with more traditional client server computing?

b) Discuss what is novel about cloud compoting as a concept?

c) Discuss any 2 of the "on demand access" services (e.g, Haas, lasS, or SaaS) that are made possible by
cloud computing for their organization
Compare and Contrast Cloud Computing with Traditional Client-Server Computing:

Cloud computing and traditional client-server computing differ in several key aspects:

1. Ownership and Management of Resources:

o In traditional client-server computing, organizations own, manage, and maintain
their hardware, software, and infrastructure on-premises.
o In cloud computing, the cloud service provider owns, manages, and maintains the
resources in off-site data centers, allowing customers to access them over the
internet.
2. Scalability and Flexibility:
o Cloud computing offers scalability on-demand, allowing users to increase or
decrease resources as needed.
o Traditional client-server setups often require upfront investments in infrastructure
that may not be easily scalable or flexible.
3. Cost Structure:
o Cloud computing operates on a pay-as-you-go or subscription model, where users
pay for the resources they consume.
o Traditional client-server models involve significant upfront costs for hardware
and software purchases.
4. Accessibility and Maintenance:
o Cloud computing provides remote access to resources from anywhere with an
internet connection.
o Traditional client-server systems may require physical presence at the data center
for maintenance and management.
5. Security and Compliance:
o Cloud providers offer robust security measures, but some organizations may have
concerns about data security in a shared environment.
o Traditional client-server setups allow for direct control over security measures but
require dedicated resources for compliance.

What is Novel About Cloud Computing as a Concept:

Cloud computing introduces several novel concepts compared to traditional IT models:

1. Resource Virtualization: Cloud computing abstracts physical hardware into virtualized

resources that can be dynamically allocated based on demand.
2. Elasticity: The ability to scale resources up or down quickly in response to changing
needs sets cloud computing apart from traditional fixed-capacity setups.
3. Self-Service Provisioning: Users can provision and manage resources independently
through web interfaces or APIs without direct IT intervention.
4. Global Accessibility: Cloud services are accessible from anywhere globally, enabling
remote collaboration and operations without geographical constraints.
5. Subscription-Based Model: Cloud computing offers a consumption-based pricing
model, allowing organizations to pay only for the resources they use rather than investing
in fixed infrastructure.
On-Demand Access Services Enabled by Cloud Computing:

1. Infrastructure as a Service (IaaS):

o IaaS provides on-demand access to computing infrastructure like servers, storage,
and networking.
o Organizations can provision and scale resources as needed without managing
physical hardware.
2. Software as a Service (SaaS):
o SaaS delivers ready-to-use applications hosted in the cloud.
o Users access applications via web browsers or clients without the need for
installation or maintenance.

1. cloud computing innovations and benefits for organizations.

c) Using an example of an organization that you are familiar with, justify why the organization
should adopt cloud computing for their organization.[8)

d) Example of an Organization: XYZ Corporation

e) 1. Cost Savings and Scalability: XYZ Corporation, a medium-sized
manufacturing company, should adopt cloud computing to benefit from cost
savings and scalability. By moving their IT infrastructure to the cloud, XYZ can
reduce capital expenditure on hardware and maintenance. They can pay for
cloud services based on their actual usage, allowing them to scale up or down as
needed without significant upfront investments. This flexibility aligns with the
company’s budgetary constraints and growth plans.
f) 2. Enhanced Collaboration and Accessibility: Cloud computing would enable
XYZ Corporation to enhance collaboration among its teams spread across
different locations. With cloud-based workflow tools and file-sharing applications,
employees can access, edit, and share documents in real-time from anywhere
with an internet connection. This increased accessibility promotes seamless
collaboration and improves overall productivity within the organization.
g) 3. Improved Security and Disaster Recovery: Security is a critical concern for
XYZ Corporation, especially regarding data protection and disaster recovery. By
migrating to the cloud, the company can benefit from robust security measures
implemented by cloud service providers. Additionally, cloud-based disaster
recovery solutions offer automated backups and data redundancy, ensuring
business continuity in case of unforeseen events such as system failures or
cyberattacks.
h) 4. Conclusion: In conclusion, XYZ Corporation stands to gain numerous
advantages by adopting cloud computing for their organization. From cost
savings and scalability to enhanced collaboration and improved security,
transitioning to the cloud aligns with the company’s strategic goals and
operational efficiency objectives.
Question 3

Explain the meaning of the following terms and give examples where appropriate:

a) Middleware

b) Persistent objects:

c) Logical clock

d) Concurrent server

e) Home-based approach to flat naming

Middleware: Middleware is software that acts as a bridge between applications and

operating systems, facilitating communication and data management. It sits between
the operating system and applications to enable data exchange and interaction.
Examples of middleware activities include handling data, API management,
authentication, and messaging services.

Persistent Objects: Persistent objects refer to objects in programming that retain

their state even after the program terminates or the system is shut down. These objects
are typically stored in a persistent storage medium like a database or file system. An
example of persistent objects could be user profiles in a social media application that
are saved in a database and remain accessible even when the application is closed.

Logical Clock: A logical clock is a concept used in distributed systems to order

events based on causality rather than real-time. It assigns logical timestamps to events
to establish an ordering without relying on physical time synchronization. An example of
a logical clock is Lamport clocks used in distributed systems to track event ordering.

Concurrent Server: A concurrent server is a type of server that can handle multiple
client requests simultaneously. It is designed to process requests concurrently, allowing
for efficient utilization of resources and improved performance. An example of
a concurrent server is a web server that can serve multiple users accessing different
web pages at the same time.

Home-Based Approach to Flat Naming: The home-based approach to flat

naming is a naming scheme used in distributed systems where each object has a
unique home location responsible for managing it. This approach simplifies object
location and access by associating each object with a specific home node. An example
 could be in a distributed file system where files are assigned to specific home nodes for
storage and retrieval.

Question 6

a) What is the difference between parallel peocessing and distributed processing?

b) Discuss the diflerences between Apache Hadoop and Apache Spark.

c) Describe a real workd problems that can be solved using Spark Streaming,

Difference Between Parallel Processing and Distributed Processing:

Parallel processing involves the simultaneous use of multiple processors to solve a

single problem, typically sharing memory and coordinating tasks to speed up
computation. On the other hand, distributed processing utilizes multiple interconnected
computers or nodes to work together on a problem, often without shared memory, with
each node having its own resources and executing tasks independently. In parallel
processing, all processors work on the same task simultaneously, while in distributed
processing, different nodes may work on different parts of the problem concurrently.

Differences Between Apache Hadoop and Apache Spark:

Apache Hadoop:

 Introduction: Hadoop is an ecosystem designed for distributed data processing,

originating from Yahoo in 2006 and later becoming an Apache open-source project.
 Components: It includes HDFS for storage, YARN for resource management, and
MapReduce for parallel data processing.
 Processing Model: Hadoop follows a batch processing model suitable for large
datasets and leverages commodity hardware for scalability.
 Advantages: Cost-effective, fault-tolerant, handles structured and unstructured data
well.
 Disadvantages: Java-based, not ideal for small data environments, potential stability
issues.

Apache Spark:

 Introduction: Spark is an open-source framework developed at UC Berkeley in 2012,

focusing on fast data processing using in-memory computing.
 Components: Spark Core for core functionalities, Spark SQL for structured data
processing, Spark Streaming for real-time data streams, MLlib for machine learning,
and GraphX for graph analytics.
 Processing Model: Supports batch, stream, interactive, machine learning, and graph
computation workloads with a focus on speed and efficiency.
 Advantages: In-memory processing for faster computation, supports various workloads
including real-time stream processing and machine learning.
 Disadvantages: Memory-intensive, may have issues with small files, fewer built-in
algorithms compared to some other tools.

Real-World Problem Solved Using Spark Streaming:

One real-world problem that can be effectively solved using Spark Streaming is real-
time fraud detection in financial transactions. By leveraging Spark’s ability to process
streaming data in near real-time combined with its machine learning capabilities from
MLlib, organizations can analyze transaction patterns as they occur. This enables the
detection of anomalies or suspicious activities instantly, allowing for immediate action to
prevent fraudulent transactions. Spark Streaming’s ability to handle high-throughput
data streams efficiently makes it a powerful tool for such time-sensitive applications.

question

Explain two benefits of middleware to distributed systems programmers, providing an

example for each benefit

a) i. Consider a distributed system consisting of four replicated servers. Each of the

servers is available at any instance with a probability of 90%. If the system is designed
so that the system can be operational if any one of the four servers is opcrational,
compute the overall system availability?(3)
 ii.What if the system is designed such that all four servers have to be available for the entire
system to be available?[3)

c) Discuss any four advantages of peer-to-peer systems?(8]

d) What is the difference between a distributed operating system and a network operating system?

a) Benefits of Middleware to Distributed Systems Programmers:

i. Simplified Communication: Middleware provides a layer of abstraction that simplifies communication

between different components in a distributed system. It hides the complexities of low-level network
protocols and provides high-level APIs for developers to easily exchange messages and data.

Example: Consider a distributed system where multiple microservices need to communicate with each
other. By using middleware like Apache Kafka or RabbitMQ, programmers can publish and subscribe to
messages using topics or queues without worrying about the underlying network details. This simplifies
the development process and makes it easier to establish communication channels between different
components.

ii. Scalability and Load Balancing: Middleware often includes features for managing load balancing and
scaling distributed applications. It can distribute incoming requests across multiple servers or instances
to ensure optimal performance and resource utilization.

Example: In a distributed e-commerce system, middleware like Nginx or HAProxy can be used to balance
the load between multiple web servers hosting the online store. It can direct user requests to the least
busy server, distributing the workload evenly and preventing any single server from becoming
overwhelmed. This helps in achieving high availability and scalability for the system.

b) System Availability Calculations:

i. If the system is designed so that the system can be operational if any one of the four servers is
operational:

- The overall system availability can be calculated using the formula for multiple redundant components
in parallel:

Total Availability = 1 - ((1 - individual component availability)^number of components)

Total Availability = 1 - ((1 - 0.9)^4) = 1 - (0.1^4) = 1 - 0.0001 = 0.9999 or 99.99%

ii. If all four servers have to be available for the entire system to be available:

- In this case, the overall system availability would be equal to the availability of the least available
component, which is 90% in this scenario.

c) Advantages of Peer-to-Peer Systems:

1. Decentralization: Peer-to-peer systems do not rely on central servers or infrastructure, allowing for
greater resilience and fault tolerance. If one peer fails, the network can still operate.

2. Scalability: Peer-to-peer systems can scale efficiently as each peer contributes resources to the
network, allowing for the expansion of services and data storage without relying on dedicated servers.

3. Anonymity and Privacy: Peer-to-peer systems offer increased privacy and anonymity as peers
communicate directly with each other without the need for centralized intermediaries, protecting user
data.

4. Reduced Costs: Peer-to-peer systems can lower costs by distributing the workload and resource
requirements among peers, reducing the need for expensive server infrastructure and maintenance.
d) Difference between Distributed Operating System and Network Operating System:

- A Distributed Operating System (DOS) is an operating system that runs on a network of computers and
manages resources across multiple nodes in the network. It provides a unified interface for users to
access resources distributed across the network.

- A Network Operating System (NOS) is an operating system specifically designed for managing network
resources and providing network services, such as file sharing, printer access, and user authentication. It
focuses on communication and coordination between devices on a network.

In summary, a DOS focuses on managing resources and coordinating tasks across a distributed system of
computers, while a NOS is focused on providing network services and communication capabilities within
a local area network (LAN) or wide area network (WAN).

Question 2

a) Compare and contrast cloud computing with more traditional client-server computing?

b) Discuss what is novel about cloud computing as a concept?

c) Discuss any 2 of the "on demand access" services (c.g.. Haas, laaS, or SaaS) that are made possible
by cloud computing.

d) Using an example of an organization that you are familiar with, justify why the organization should
adopt cloud computing for their organization.(8)

a) Compare and Contrast Cloud Computing with Traditional Client-Server

Computing:

Cloud computing and traditional client-server computing are two different paradigms in
the realm of information technology. In traditional client-server computing, applications
and data are hosted on a dedicated server within an organization’s premises. This setup
requires the organization to manage and maintain the hardware, software, security, and
backups associated with the server. Users access these resources through client
applications installed on their devices.

On the other hand, cloud computing involves delivering services over the internet.
Instead of hosting applications and data on local servers, cloud computing relies on
remote servers hosted by third-party providers. Users access these services over the
internet using web browsers or lightweight client applications. Here are some key points
of comparison between the two:
1. Scalability: Cloud computing offers greater scalability compared to traditional client-
server models. With cloud services, organizations can easily scale up or down based on
their needs without having to invest in additional infrastructure.

2. Cost: Cloud computing often follows a pay-as-you-go model, where organizations pay
for the resources they use. This can be more cost-effective than maintaining and
upgrading on-premises servers in traditional client-server setups.

3. Maintenance: In traditional client-server computing, organizations are responsible for

maintaining and updating their own servers. In contrast, cloud computing providers
handle the maintenance, updates, and security of the underlying infrastructure in a
cloud environment.

4. Accessibility: Cloud computing enables users to access applications and data from
anywhere with an internet connection, promoting remote work and collaboration.
Traditional client-server setups may require users to be connected to a specific network
to access resources.

b) Novelty of Cloud Computing as a Concept:

One of the key novel aspects of cloud computing is its abstraction of infrastructure from
end-users. Users no longer need to worry about the physical hardware or underlying
technical details when utilizing cloud services. The concept of virtualization plays a
significant role in cloud computing by allowing resources to be dynamically allocated
based on demand.

Moreover, cloud computing introduces the idea of elasticity, where resources can
automatically scale up or down based on workload requirements. This flexibility enables
organizations to optimize resource utilization and cost efficiency.

c) On-Demand Access Services Enabled by Cloud Computing:

1. Infrastructure as a Service (IaaS): IaaS provides virtualized computing resources over

the internet, including virtual machines, storage, and networking capabilities. Users can
provision and manage these resources on-demand, paying only for what they use.
Examples of IaaS providers include Amazon Web Services (AWS), Microsoft Azure,
and Google Cloud Platform.

2. Software as a Service (SaaS): SaaS delivers software applications over the internet on
a subscription basis. Users can access these applications through web browsers
without needing to install or maintain them locally. Popular examples of SaaS include
Salesforce CRM, Microsoft Office 365, and Google Workspace.

d) Justification for Adopting Cloud Computing for an Organization:

 An organization that should consider adopting cloud computing is XYZ Company, a
medium-sized retail business expanding its operations globally. By transitioning to
cloud-based solutions, XYZ Company can benefit from:

1. Scalability: As XYZ Company expands into new markets, cloud computing allows them
to easily scale their IT infrastructure to support increased demand without significant
upfront investments in hardware.

2. Cost Efficiency: With fluctuating demand in different regions, XYZ Company can
leverage the pay-as-you-go model of cloud services to optimize costs based on usage
patterns, avoiding over-provisioning of resources.

3. Global Accessibility: Cloud computing enables XYZ Company’s employees across

different locations to access centralized data and applications securely from anywhere,
fostering collaboration and efficiency in a distributed work environment.

By embracing cloud computing, XYZ Company can enhance its agility, competitiveness,
and operational efficiency while focusing on its core business objectives.

Or a) Comparison of Cloud Computing with Traditional Client-Server Computing:

- In traditional client-server computing, data and applications are stored and run on dedicated servers
within an organization's premises, requiring physical maintenance and IT resources. In contrast, cloud
computing relies on remote servers hosted by third-party providers, allowing organizations to access
services and resources over the internet.

- Traditional client-server models often require significant upfront costs for hardware and software
procurement, while cloud computing offers a pay-as-you-go model, where organizations only pay for the
resources they consume.

- Cloud computing provides scalability and flexibility, allowing organizations to easily scale resources up
or down based on demand, while traditional client-server setups may require additional hardware or
infrastructure upgrades to accommodate growth.

b) Novelty of Cloud Computing:

- One novel aspect of cloud computing is the concept of virtualization, where physical hardware
resources are abstracted and pooled together, allowing for efficient utilization and allocation of
resources.

- Another novel aspect is the self-service provisioning model, where users can easily request and deploy
resources without needing direct intervention from IT administrators. This on-demand access to
resources promotes agility and flexibility within organizations.

c) On-Demand Access Services Enabled by Cloud Computing:

i. IaaS (Infrastructure as a Service): With IaaS, organizations can access virtualized computing resources
such as virtual machines, storage, and networking on a pay-as-you-go basis. Users can quickly provision
and scale these resources based on their requirements without investing in physical hardware.

ii. SaaS (Software as a Service): SaaS allows organizations to access and use software applications hosted
on the cloud. Users can access these applications through a web browser without the need for
installation or maintenance. Examples of SaaS include Office 365, Salesforce, and Google Workspace.

d) Justification for Adoption of Cloud Computing for an Organization:

- Example Organization: XYZ Company is a medium-sized manufacturing company with multiple

locations and a growing need for digital collaboration and data storage.

- Justification: XYZ Company should adopt cloud computing for their organization to leverage the
benefits of scalability, cost-efficiency, and improved collaboration.

- By migrating their data and applications to the cloud, XYZ Company can easily scale their resources
based on demand and reduce upfront costs associated with maintaining physical servers.

- Cloud-based collaboration tools and services can enhance communication and productivity among
employees in different locations, promoting better teamwork and efficiency.

- Additionally, cloud-based data storage and backup solutions can provide increased security, reliability,
and accessibility to critical business information, ensuring business continuity and disaster recovery
capabilities for XYZ Company.

Question 3

a) Arianne is the tech lead at XCheck and has been tasked with creating a distributed database for
accounts and payments. In order to support millions of customers, the data of different accounts will
live on different servers. Arianne is trying to decide between the Two Phase Commit protocol (2PC)
and the Paxos consensus protoco! (laxos).

i) For 2PC, assume that there are two participating servers A and B, and both send "VoteCommit" to
the coordinator. The coordinator. sends out the"DoCommit" message to A. but crashes immediately
before sending a message to B. When do A and B commit their respective changes to their local
database?

ii) For Paxos, Arianne is concerned about having to wait for a majority of acceptors. She considers an
implementation of Paxos where a proposer waits for less than a majority of acceptors to answer OK to
a Prepare or an Accept message before it proceeds to executing the next steps. Describe a scenario
where Arianne's implementation may not correctly match Paxes' service guarantees[5]

For Paxes, calculate the number of messages that will be exchanged to update the balance of a single
account? A ssume that there are 100 servers
b) Unnecessary view changes (or elections/scouts) can prevent consensus from making progress.
Why? Propose a mechanism to prevent unnecessary view changes in practice.

a) Distributed Database Protocols Comparison:

i) Two Phase Commit protocol (2PC):

- In the scenario provided, if the coordinator crashes immediately after sending the "DoCommit"
message to server A but before sending a message to server B, server A will commit its changes to its
local database since it received the "DoCommit" message. However, server B will not commit its
changes as it did not receive any instruction from the coordinator.

ii) Paxos Consensus Protocol:

- In a scenario where a proposer waits for less than a majority of acceptors to answer OK to a Prepare or
an Accept message before proceeding to the next steps, there is a risk of violating Paxos' service
guarantees. If the proposer proceeds before receiving responses from a majority, it may lead to
inconsistencies and potential conflicts in the consensus process, impacting the correctness of the
consensus outcome.

For Paxos, calculate the number of messages exchanged to update the balance of a single account with
100 servers:

- In Paxos, the number of messages exchanged to update the balance of a single account would involve
multiple rounds of Prepare, Accept, and Commit messages exchanged between the proposer and
acceptors. The specific calculation of the exact number of messages exchanged would depend on the
implementation details and the number of rounds required to achieve consensus.

b) Preventing Unnecessary View Changes in Consensus:

- Unnecessary view changes, also known as leader changes in distributed systems, can hinder consensus
progress by introducing delays and disruptions in the decision-making process.

- One mechanism to prevent unnecessary view changes is to implement a heartbeat mechanism where
the leader periodically sends heartbeat messages to all nodes in the system to indicate its liveness and
availability. If a node does not receive a heartbeat from the leader within a certain interval, it can trigger
a view change process to elect a new leader.
- Additionally, introducing a threshold or quorum-based approach can help prevent premature view
changes by ensuring that a certain number of nodes acknowledge the need for a view change before
proceeding with the transition to a new view.

- By carefully monitoring the leader's activity, implementing timeout mechanisms, and considering a
consensus-based approach for initiating view changes, unnecessary view changes can be minimized,
ensuring the stability and progress of the consensus protocol in practice.

Question 4 €

a) Resource sharing is the main motivating factor for constructing distributed systems.

Resources such as printers, files, web pages or database records are managed by servers of the
appropriate type. The construction of distributed systems produces many challenges. Discuss any of
the five main challenges and how they can be addressod.(10]

b) Explain what is meant by a virtual organization and give a hint on how such organizations could be
implemented.[5]

c) A user arrives at a railway station that he has never visited before, carrying a mobile device that is
capable of wireless networking. The user could be provided with information about the local services
and amenities at that station, without entering the station's name or attributes. Make use of this
scenario to describe the features of distributed pervasive systems.[10)

a) Challenges in Distributed Systems and Addressing Them:

i) Scalability: As the number of nodes and users in a distributed system grows, scalability becomes a
critical challenge. To address this, techniques such as load balancing, sharding, and vertical/horizontal
scaling can be implemented. Load balancing distributes incoming requests among multiple servers to
optimize resource usage. Sharding involves partitioning the data across multiple servers to distribute the
workload evenly. Vertical scaling refers to increasing the resources of individual nodes, while horizontal
scaling involves adding more nodes to the system.

ii) Fault Tolerance: Distributed systems are prone to failures, leading to data inconsistencies and service
disruptions. To ensure fault tolerance, mechanisms such as replication, redundancy, and fault detection
can be employed. Replication involves creating copies of data across multiple nodes to ensure data
availability in case of failures. Redundancy involves having backup systems or components to take over
in case of failures. Fault detection mechanisms monitor the system for failures and trigger recovery
actions automatically.

b) Virtual Organization:
A virtual organization is a network of geographically dispersed individuals or groups linked together
through information and communication technologies to achieve common goals. These organizations
operate without physical boundaries, utilizing virtual teams and platforms for collaboration and
communication. Implementing virtual organizations involves leveraging tools such as virtual meeting
software, project management platforms, and cloud-based collaboration tools. Additionally, establishing
clear communication channels, setting goals and expectations, and fostering a culture of trust and
accountability are essential elements of virtual organizations.

c) Features of Distributed Pervasive Systems in Railway Station Scenario:

- Context Awareness: The distributed pervasive system at the railway station can utilize sensors and
location-based services to gather information about the user's surroundings and context. This
information can include the user's location, the time of day, available services, and amenities at the
station.

- Seamless Connectivity: The mobile device's wireless networking capabilities allow the user to access
information about local services and amenities without the need to explicitly enter any details. The
distributed pervasive system enables seamless connectivity and information retrieval based on the
user's context.

- Ubiquitous Access: The distributed pervasive system ensures that the user can access relevant
information about the railway station's services and amenities from anywhere within the station
premises. This ubiquitous access enhances the user experience and provides convenience.

- Real-time Updates: The distributed system can provide real-time updates and notifications to the user
about train schedules, platform changes, delays, and other relevant information. This ensures timely and
accurate information delivery to enhance the user's overall experience at the railway station.

Question 5

a) Describe the Nakamoto consensus.(5)

b) Discuss the improvements made to the Conflux decentralized blockchain system to solve some of
the limitations of the Nakamoto consensus.(10]

c) How would you use blockchain to design a system to solve a real world problem?| 10]

a) Nakamoto Consensus:

The Nakamoto Consensus, also known as Proof of Work (PoW), is a consensus mechanism used in
blockchain technology to validate transactions and secure the network. In the Nakamoto Consensus,
miners compete to solve complex mathematical puzzles in order to add new blocks to the blockchain.
The first miner to solve the puzzle broadcasts the solution to the network, and if the solution is verified
by other nodes, the new block is added to the blockchain. This process requires significant
computational power and energy to deter malicious actors from manipulating the blockchain.

b) Improvements in Conflux Decentralized Blockchain System:

Conflux is a decentralized blockchain system that has made several improvements to address limitations
of the Nakamoto Consensus, including:

i) Scalability: Conflux uses a novel parallelized architecture that allows for faster transaction processing
and higher throughput compared to traditional blockchain systems. By running consensus algorithms in
parallel, Conflux can achieve higher scalability without sacrificing security.

ii) Low Latency: Conflux utilizes a Tree Graph structure that enables faster block propagation and
confirmation times. This reduces latency in transaction processing and improves the overall efficiency of
the blockchain network.

iii) Fairness: Conflux incorporates a decentralized governance model that ensures fairness in decision-
making processes within the network. This allows for community-driven governance and prevents
centralization of power.

iv) Security: Conflux integrates advanced cryptographic techniques and security measures to protect
against attacks and ensure the integrity of the blockchain. By incorporating robust security mechanisms,
Conflux enhances the overall security of the decentralized system.

c) Designing a Blockchain System to Solve a Real-World Problem:

To use blockchain to solve a real-world problem, such as supply chain transparency, the following steps
can be taken:

i) Identify the problem: Determine the specific issue within the supply chain that blockchain can address,
such as tracking the origin of products, ensuring product authenticity, or improving traceability.
ii) Design the system: Develop a blockchain system that includes smart contracts to automate and
enforce transparency in the supply chain. Use a permissioned blockchain to control access and
permissions based on the different stakeholders involved.

iii) Implement tracking mechanisms: Utilize IoT devices, QR codes, or RFID tags to track the movement of
products along the supply chain. This data is then recorded on the blockchain to create an immutable
record of the product's journey.

iv) Establish trust and transparency: Allow stakeholders to access the blockchain to view real-time
updates on product movements, verify authenticity, and ensure compliance with regulations. This
transparency enhances trust among participants and improves the overall integrity of the supply chain.

v) Continuous monitoring and improvement: Regularly monitor the blockchain system, analyze data, and
make adjustments as needed to optimize efficiency, accuracy, and transparency within the supply chain.
By continuously improving the system, the real-world problem can be effectively addressed using
blockchain technology.

Question 7

a) Discuss the differences between the following two data-centric consistency models: sequential
consistency and entry consistency?

* b) Explain the differences between sequential and casual and entry consistency. Which of these
models would be better, from the performance point of view, for typical business applications
operating on sets of data?[8]

c) Justify the kind of protocols for delivery of updates to replicas you would choose to achieve
sequential consistency.

d) Discuss the kind of consistency that is provided for replicated databases if all operation requests
are delivered by totally ordered multicast.

Differences Between Sequential Consistency and Entry Consistency

Sequential consistency and entry consistency are two data-centric consistency models
in distributed systems that govern how data is accessed and updated across multiple
nodes. Here are the key differences between these two models:
1. Definition:

 Sequential Consistency: In the sequential consistency model, the result of any

execution is as if all read and write operations by all processes were executed in some
sequential order. Each process’s operations appear in this sequence as specified by its
program.
 Entry Consistency: Entry consistency, on the other hand, focuses on ensuring that all
replicas of an object are updated in the same order. It guarantees that updates to a
single object are applied in the same order at all replicas.

2. Scope:

 Sequential Consistency: Sequential consistency applies to the overall order of

operations across all processes in a distributed system.
 Entry Consistency: Entry consistency specifically targets the ordering of updates for a
single object replicated across multiple nodes.

3. Granularity:

 Sequential Consistency: This model enforces a global order of operations, ensuring

that all processes see the same sequence of events.
 Entry Consistency: Entry consistency operates at a finer granularity, focusing on
maintaining the order of updates for individual objects.

4. Guarantees:

 Sequential Consistency: Ensures that all processes observe a consistent global

ordering of operations.
 Entry Consistency: Guarantees that updates to a single object are applied in the same
order at all replicas.

5. Complexity:

 Sequential Consistency: Implementing sequential consistency can be more complex

due to the need to coordinate and synchronize operations across all nodes.
 Entry Consistency: Entry consistency may be simpler to implement as it deals with
maintaining order within a specific object’s updates.

6. Performance Impact:

 Sequential Consistency: Achieving sequential consistency may introduce higher

latency and coordination overhead, impacting performance.
 Entry Consistency: Entry consistency, focusing on individual objects, may offer better
performance as it involves fewer synchronization requirements across nodes.

Differences Between Sequential, Causal, and Entry Consistency

1. Sequential Consistency vs. Causal Consistency:

 Sequential consistency enforces a global order of operations across all processes, while
causal consistency ensures that causally-related shared accesses are seen in the same
order.

2. Causal Consistency vs. Entry Consistency:

 Causal consistency maintains causal relationships between shared accesses but does
not require a total order, whereas entry consistency focuses on maintaining update
order for individual objects.

Performance Considerations for Business Applications

From a performance perspective for typical business applications operating on sets of

data, entry consistency may be preferable due to its focus on maintaining the update
order for individual objects. This targeted approach can lead to lower coordination
overhead and potentially better performance compared to achieving strict sequential
consistency which requires enforcing a global order of operations across all processes.

Protocols for Achieving Sequential Consistency

To achieve sequential consistency in distributed systems, protocols such as Total Order

Broadcast (TOB) or Total Order Multicast (TOM) can be employed. These protocols
ensure that messages are delivered and processed in the same total order by all nodes
in the system, thereby enforcing a consistent global ordering of operations.

Consistency Provided for Replicated Databases with Totally Ordered Multicast

If all operation requests are delivered by totally ordered multicast in replicated

databases, strict sequential consistency would be achieved. This means that all replicas
would see and process operations in the exact same total order, ensuring a consistent
view of data across all nodes.

a) Differences between Sequential Consistency and Entry Consistency:

- Sequential Consistency: In sequential consistency, the order of operations on a single object must be
the same for all processes in the system. This means that each process sees the operations in the same
order with respect to that object.
- Entry Consistency: Entry consistency is a weaker form of consistency where the order of operations
only needs to be consistent within an object (entry). This means that different processes can have
different views of the order of operations for different objects, as long as the order is consistent within
each object.

b) Differences between Sequential, Causal, and Entry Consistency:

- Sequential Consistency: Requires all processes to agree on the order of operations on all objects. It
ensures that all processes see the same order of operations, which can lead to delays and inefficiencies
in systems with high contention.

- Causal Consistency: Focuses on preserving causality relationships between operations. It allows for
some operations to be seen out of order as long as there is no causal dependency between them. This
can improve performance in systems where strict ordering is not necessary.

- Entry Consistency: Provides consistency within individual objects (entries) but allows for different
views of the order of operations across objects. This can be more efficient for systems where operations
on different objects do not need to be strictly ordered.

From a performance point of view, Entry Consistency would likely be better for typical business
applications operating on sets of data. This is because it offers more flexibility and can reduce
contention and delays by allowing different views of the order of operations for different objects.

c) Protocols for Achieving Sequential Consistency:

To achieve sequential consistency in a distributed system, protocols such as Total Order Broadcast (TOB)
or Atomic Broadcast can be used. Total Order Broadcast ensures that all messages are delivered to all
processes in the same order, which helps maintain the order of operations and achieve sequential
consistency across the system. By enforcing a total order of message delivery, the system can guarantee
that all processes see the operations in the same order, leading to sequential consistency.

d) Consistency in Replicated Databases with Totally Ordered Multicast:

If all operation requests are delivered by totally ordered multicast, the replicated databases would
achieve strong consistency. Totally ordered multicast ensures that all updates are delivered in the same
order to all replicas, preventing inconsistencies and ensuring that all replicas are up-to-date and
consistent at all times. This level of consistency ensures that all operations are performed in the same
order across all replicas, maintaining data integrity and avoiding conflicts.

Question 1

Facebook is a distributed system discuss[25]

Facebook is a massive social networking platform that operates as a distributed system. A distributed
system is a collection of independent nodes that work together to provide a unified service. In the case
of Facebook, the distributed system architecture is crucial for handling the enormous amount of data,
users, and interactions on the platform.

Here are some key aspects of Facebook as a distributed system:

1. **Scalability:** One of the primary reasons for using a distributed system architecture is scalability.
Facebook serves billions of users worldwide, and a centralized system would struggle to handle the
volume of data and requests. By distributing its workload across multiple servers and data centers,
Facebook can efficiently scale its infrastructure to meet the demand.

2. **Fault Tolerance:** Distributed systems are designed to be resilient to failures. If a single server or
data center goes down, the distributed system can continue to function without significant disruption.
Facebook utilizes redundancy and replication techniques to ensure that user data is safe and accessible
even in the event of hardware failures.

3. **Data Replication:** To provide users with a seamless experience, Facebook employs data
replication across multiple servers. This ensures that user data remains consistent and available across
different regions. By replicating data, Facebook can improve performance and reduce latency for users
accessing the platform from different parts of the world.

4. **Load Balancing:** Facebook uses load balancing techniques to distribute incoming traffic evenly
across its servers. This ensures that no single server is overwhelmed with requests, optimizing
performance and preventing bottlenecks. Load balancers help Facebook handle spikes in traffic and
maintain a responsive user experience.

5. **Caching:** To improve performance and reduce the load on backend servers, Facebook utilizes
caching. Commonly accessed data and content are cached at edge servers or CDN (Content Delivery
Network) nodes to minimize latency and deliver content quickly to users. Caching plays a vital role in
enhancing the user experience on the platform.

6. **Consistency and Availability:** Maintaining consistency and availability in a distributed system like
Facebook is a complex challenge. Facebook employs consistency models that balance data integrity with
performance and availability requirements. Techniques such as eventual consistency and quorum-based
systems are used to ensure that data remains consistent across distributed nodes.

7. **Networking Infrastructure:** Facebook invests heavily in its networking infrastructure to support

its distributed system. High-speed networking, data centers located strategically around the globe, and
sophisticated routing protocols enable Facebook to deliver content quickly and reliably to users
worldwide.

In conclusion, Facebook's distributed system architecture plays a crucial role in delivering a seamless
user experience to its global user base. By leveraging scalability, fault tolerance, data replication, load
balancing, caching, and robust networking infrastructure, Facebook can efficiently handle the massive
scale of data and interactions on its platform.

Question 2

a) Compare RPC and RMI with respect to implementation.(10)

b) Explain how logical clocks are implemented in a distributed system.(6)

c) Let BETTR be a company that promises to take a user's photos and apply a collection of
independent filters to each photo so as to improve its quality (according to some proprietary model of
quality, eg., one that prefers sharper, brighter, warmer-toned photos). Assume that the estimated
demand for this service is very large (e.g.,millions of photos per hour).

i) Assume you have just been hired by BETTR and in your first technical meeting, your boss asks you
whether you think it is a good idea to use a master-workers parallelization pattern to process this
workload and why. What answer would you give her?

ii) Your boss then says that someone told her that replication is an important strategy for improved
availability. She asks you to explain what is meant by this and whether there are drawbacks to using
replication in the case of BETTR's photo processing pipeline. What answer would you give her?

a) Comparison of RPC and RMI with respect to implementation:

1. **RPC (Remote Procedure Call):**

 - Implementation: In RPC, a client program can call procedures on a remote server as if they were local
procedures. This is achieved through stubs and marshalling. The client-side stub converts the procedure
call into network communication, sends it to the server, which then executes the procedure, sends the
result back, and the stub on the client side unmarshals the result.

2. **RMI (Remote Method Invocation):**

- Implementation: RMI is specific to Java and enables objects to invoke methods on remote objects. It
uses Java interfaces and classes to define remote objects and client applications. RMI utilizes Java's
serialization to pass objects by value between the client and server, and stubs play a crucial role in RMI
implementation.

**Key Differences:**

- RPC is more language-neutral while RMI is Java-specific.

- RMI provides support for passing objects by reference but RPC generally deals with passing
parameters by value.

- In terms of implementation, both RPC and RMI rely on stubs, but RMI typically has a higher level of
abstraction due to its object-oriented nature.

b) Implementation of Logical Clocks in a Distributed System:

In a distributed system, logical clocks help order events that occur in different processes. The most
common implementation of logical clocks is based on Lamport timestamps, which assign a unique
timestamp to each event in the system. Here's how logical clocks are implemented:

1. Each process in the distributed system maintains its logical clock, initially set to zero.

2. When an event occurs at a process, the process increments its logical clock by 1 and assigns the
incremented value to the event.

3. When a process sends a message, it includes its current logical clock value in the message.

4. Upon receiving a message, the receiving process updates its logical clock to be the max of its current
logical clock value and the received timestamp + 1.

5. This way, logical clocks help in establishing the ordering of events in a distributed system.

c)
i) Using Master-Workers Parallelization Pattern for BETTR's Photo Processing Workload:

- The Master-Workers parallelization pattern involves a master node distributing tasks to worker nodes,
which process the tasks concurrently. In the case of processing millions of photos per hour, utilizing this
pattern can significantly improve the efficiency and speed of processing by leveraging parallel computing
capabilities.

- I would recommend using the Master-Workers pattern for BETTR's photo processing workload as it
allows for horizontal scaling and efficient distribution of tasks. By dividing the workload among multiple
worker nodes, BETTR can achieve faster processing times and handle the large demand effectively.

ii) Replication Strategy for Improved Availability in BETTR's Photo Processing Pipeline:

- Replication involves creating copies of data or services across multiple nodes to enhance availability
and fault tolerance. In the case of BETTR's photo processing pipeline, replicating critical components
such as the image processing algorithms, databases, and workload distribution mechanisms can improve
reliability and ensure continuous operation in the event of node failures.

- However, there are potential drawbacks to using replication in this scenario. Maintaining consistency
among replicated data or services can be challenging, especially in a high-demand environment like
photo processing. Synchronization overhead, network latency, and complexities in handling concurrent
updates across replicas are factors that need to be carefully considered when implementing replication
in BETTR's system.

Overall, while replication can enhance availability, it requires careful planning and management to
address potential drawbacks and ensure the overall stability and performance of BETTR's photo
processing pipeline.

Question 3

It is now common for people to make up their own holiday packages by booking flights, hotels and
excursions at different online sites rather than choosing a pre-determined package offered by travel
agents. When booking these independently, however, a problem exists that makes it difficult to
guarantee that you get all the components you want for your holiday. For example, by the time you
finish booking your flight, it may be that the hotel you prefer no longer has any rooms, or that the
theatre event you must see has sold out. This leaves the unpleasant possibility of being left with a
flight to a city you no longer want to visit. A proposal has been put forward to set up an online
business called allornothing.com to overcome this problem for users. The idea is that you select all the
components you want for your holiday on one single site, which then guarantees to keep them all on
hold until you come to make your final payment. If any one component fails to be booked for some
reason, the site guarantees that none of the components are booked / paid for. This way you end up
with nothing, or with exactly the bookings you want.

In the context of this proposal, discuss the following:

a) Identify issues relating to 'deadlock' that exist in this system.

b) Explain how the issues identified in 5a above can be avoided.

a) Issues relating to 'deadlock' in the proposed system:

In the scenario described, where all components of a holiday package must be successfully booked or
none at all, there is a potential for deadlock to occur. Deadlock is a situation in which two or more
processes are unable to proceed because each is waiting for the other to release a resource, ultimately
leading to a standstill. In the context of allornothing.com, the following deadlock scenarios may arise:

1. **Component Dependencies:** If different components of the holiday package have dependencies

on each other, such as a certain flight being required to book a specific hotel room, there is a risk of
deadlock if one component cannot be booked. For example, if the flight booking fails but the hotel
booking is already reserved, preventing any further progress.

2. **Limited Availability:** Deadlock may occur if all the desired components have limited availability
and are in high demand. If one component gets booked but the others are not available, the system may
be unable to proceed with the booking of the remaining components, leading to a deadlock situation.

3. **Reservation Durations:** If components are kept on hold for a specific duration while waiting for
final payment, deadlock can occur if the hold periods overlap and users are unable to finalize their
bookings within the specified time frame, causing all components to be released without confirmation.

b) How to avoid deadlock in the proposed system:

To prevent deadlock in the system proposed by allornothing.com, the following strategies can be
implemented:

1. **Timeout Mechanism:** Implement a timeout mechanism that releases the hold on components if
the final payment is not made within a specified time frame. This will prevent indefinite waiting and
ensure that components are released for other users if the booking process stalls.
2. **Booking Priorities:** Establish a prioritization system for booking components based on
dependencies and availability. Ensure that critical components are booked first to reduce the risk of
deadlock due to dependencies on unavailable resources.

3. **Conditional Bookings:** Allow users to specify alternative options for each component in case their
first choice is not available. This flexibility can help avoid deadlock by enabling the system to proceed
with alternative bookings if the primary choices are unavailable.

4. **Parallel Processing:** Enable concurrent processing of component bookings to reduce the waiting
time and increase the chances of successfully booking all components. Parallel processing can help avoid
situations where one component blocks the progress of others.

5. **Dynamic Resource Allocation:** Continuously monitor availability and dynamically adjust resource
allocations to maximize the chances of successful bookings. This adaptive approach can help mitigate
deadlock risks by proactively managing resource contention.

By incorporating these strategies into the design and implementation of the allornothing.com system,
the likelihood of deadlock occurring can be minimized, ensuring a smoother and more reliable booking
process for users.

Question 6

a) Assuming you are a software engineer at Goodlet, a web-based company whose main business
process relies on aggregating adverts of properties for sale. These adverts come from a large number
of estate agents. Goodlet makes money by charging each estate agent a fee every time that a Goodlet
website user converts into a prospective customer (say, by requesting further details on a property, or
by requesting to visit a property, etc.).

i) Jack, a colleague of yours, has argued that, from the viewpoint of the estate agents, the main reason
for them to participate in the distributed system created by Goodlet, is non-functional.

State whether or not you agree with Jack and explain why.(6)

il) Jill is another colleague of yours. She has argued that Goodlet should expand its business processes
by developing a RESTful API. This would allow Goodlet to charge third-party applications (say, interior
decoration companies, etc.) for access to the estate agent data they hold. Goodlet would then pay the
estate agents a share of the revenue that had been generated in this way.
Explain why this new initiative would characterize, from the viewpoint of the estate agents, a
functional reason to participate in the distributed system created by Goodlet. [6)

iii) Assume that Goodlet adopted Jill's proposal (in 3(a)ii above) for exposing a RESTful API for third-
party applications to access their data. Argue that, in this extended Goodlet system, all the
components have the features required in the definition of a distributed system.[8)

b) Explain synchronization with respect to distributed systems.(5]

a)

i) Non-functional reason for estate agents to participate in the distributed system:

- I disagree with Jack as the main reason for estate agents to participate in the distributed system
created by Goodlet is functional rather than non-functional. Estate agents participate in the system
primarily to generate leads and potential customers for their properties, which directly impacts their
business and revenue. By reaching a wider audience through Goodlet's platform, estate agents have the
opportunity to increase the visibility of their properties and attract more potential buyers, ultimately
leading to potential sales and profits.

ii) Functional reason for estate agents to participate in the distributed system with a RESTful API:

- By developing a RESTful API to provide access to estate agent data for third-party applications, Goodlet
opens up a new revenue stream by charging these third-party applications for accessing the valuable
property information. This initiative offers a tangible benefit to estate agents as they stand to receive a
share of the revenue generated through this additional channel. Participating in the distributed system
not only helps estate agents reach potential customers but also enables them to generate additional
income through revenue sharing, which is a functional reason for their continued participation.

iii) Features of a distributed system in the extended Goodlet system:

- In the extended Goodlet system with a RESTful API for third-party applications, all components exhibit
the essential features of a distributed system:

1. **Multiple Components:** The system consists of multiple components such as the Goodlet
platform, estate agent databases, third-party applications, and the RESTful API.

2. **Interconnectivity:** These components communicate and interact with each other over a
network, exchanging data and requests through the RESTful API.

3. **Concurrency:** The system handles concurrent requests from multiple users and third-party
applications accessing estate agent data simultaneously.

4. **Fault Tolerance:** The system is designed to gracefully handle failures or disruptions, ensuring
that the availability and reliability of services are maintained.
b) Synchronization in Distributed Systems:

Synchronization in distributed systems refers to the coordination and control of concurrent processes or
components to ensure consistency and avoid conflicts. In the context of distributed systems, where
multiple entities may operate independently and asynchronously, synchronization mechanisms are
crucial for maintaining data integrity and consistency. Some key aspects of synchronization in
distributed systems include:

1. **Consistency Management:** Synchronization mechanisms are used to manage data consistency

across distributed nodes. Techniques like distributed locking, versioning, and distributed transactions
help maintain a consistent view of data across the system.

2. **Concurrency Control:** Distributed systems often have multiple processes accessing shared
resources concurrently. Synchronization techniques such as mutual exclusion (e.g., using locks or
semaphores) are employed to ensure that only one process accesses a critical resource at a time,
preventing conflicts and data corruption.

3. **Coherence in Caching:** In distributed caching systems, synchronization mechanisms ensure cache

coherence by propagating updates and invalidations to all nodes that have copies of the cached data.
This avoids inconsistencies that may arise due to stale or outdated data in caches.

4. **Event Ordering:** Synchronization is essential for establishing a consistent order of events in

distributed systems. Logical clocks, timestamps, and consensus algorithms help ensure that events are
ordered correctly, even in the presence of network delays and communication issues.

Overall, synchronization plays a critical role in distributed systems by managing concurrent access,
maintaining consistency, and facilitating smooth collaboration among distributed components.

Question 5

a) Imagine you are a software engineer, in a company that is planning a move towards using web-
based distributed computing. The technical manager has called for a discussion of the challenges
involved. The fact that the existing systems wereoriginally designed with a view to being CPU-efficient
causes her particular concern.
She asks you to comment on whether her concerns are well-founded. State whether you agree with
your technical manager and briefly explain why.[8)

b) Explain why peer-to-peer architectures (P2P) are said to lead naturally to balanced loads and
graceful scaling.[6]

c) Consider the Two Generals problem. Defend the assertion that it is impossible to design an
algorithm that is guaranteed to lead to a coordinated solution.[6

d) Describe the benefits of adopting grid computing for a university(5)

a) The concerns of the technical manager regarding the move towards using web-based distributed
computing are indeed well-founded, especially considering the fact that the existing systems were
originally designed with a focus on CPU efficiency. In a distributed computing environment, emphasis is
often placed on factors such as network latency, data transfer bandwidth, and parallel processing
capabilities rather than just CPU efficiency. As such, the transition to distributed computing may require
significant re-architecting and optimization of existing systems to ensure they can effectively operate in
a distributed environment. Adapting to distributed computing may involve redesigning algorithms, data
structures, and communication protocols to account for the complexities introduced by distributed
systems, which may differ from the CPU-centric design principles of the existing systems.

b) Peer-to-peer (P2P) architectures are said to lead naturally to balanced loads and graceful scaling due
to the decentralized nature of the network. In a P2P system, each node (peer) in the network has equal
capabilities and can act as both a client and a server, contributing resources and participating in the
network's operations. This leads to several benefits:

- **Load Balancing:** In a P2P network, tasks and data can be distributed across multiple nodes,
automatically balancing the load and avoiding bottlenecks on specific nodes. Each peer can contribute
resources and share the workload, leading to a more balanced and efficient system.

- **Scalability:** P2P networks can scale gracefully as new nodes can easily join or leave the network
without affecting the overall system's performance. The distributed nature of P2P systems allows them
to handle increasing loads and accommodate a growing number of participants without centralized
constraints.

c) The Two Generals problem is an unsolvable issue in distributed computing where two
communication-restricted generals must agree on a coordinated attack plan. The assertion that it is
impossible to design an algorithm guaranteed to lead to a coordinated solution stems from the inherent
limitations of asynchronous communication and the impossibility of achieving absolute certainty in
distributed systems. In scenarios like the Two Generals problem, where messages can be lost, delayed,
or delivered out of order, there is no foolproof method to ensure that both parties have reached a
definitive agreement without any ambiguity or risk of failure. As a result, designing an algorithm that
guarantees a coordinated solution in the face of these challenges is fundamentally impossible.
d) Benefits of adopting grid computing for a university:

- **Resource Optimization:** Grid computing allows universities to efficiently utilize and share
resources such as computing power, storage, and research data across departments and research
projects. This optimizes resource usage and maximizes productivity.

- **Collaboration and Research:** Grid computing facilitates collaboration among researchers, enabling
them to access shared resources and work together on complex scientific and academic projects. This
enhances research capabilities and accelerates progress in various fields.

- **Scalability:** Grid computing provides scalability to accommodate growing computational needs

and handle large-scale simulations, data analysis, and processing requirements. Universities can easily
scale up their computing infrastructure to meet evolving demands.

- **Cost Efficiency:** By sharing resources and infrastructure through grid computing, universities can
reduce costs associated with maintaining and managing individual computing systems. This cost
efficiency allows institutions to allocate resources effectively to support research and academic
activities.

- **High Performance Computing:** Grid computing offers universities access to high-performance

computing capabilities, allowing them to tackle computationally intensive tasks and simulations that
require substantial computational power. This enables researchers to carry out advanced analyses and
simulations efficiently.

Question 1

Case Study: Google Lie. Read the passage below and antwer the questions that follow.

The mission statement of Google is "to organise the work's information and make it universally
accessible and useful." The company was bern out of an internet search research project at Stanford,
and has diversified into cloud computing, in its initial production system In 1998, Larry Page and
Serpey Brin run the company from a excage supporting up to 5 billion searches per day which
averages to about 50,000 searches per second. The main search engine, which is presented to the end
user in the form of a web page, has never experienced an outage and on average users receive query
results in 0.2seconds. Through crawling, indexing, and ranking functions, their search engine provides
one of the mont appealing search experience on the web.

Some of the services they offer include a search engine, GMail, Google Docs, Google Drive, Herols,
Google App Engine (GAE), Google Compute Engine (GCT) and Google Kubernetes Engine (GKE), GCE
gives the ablity to create virtual machines whist allocating CPU, memory, the kind of storage together
with the amount of storage. GIE allows customers to easily run their Docker containers in a fully
managed Kubernetes environment. For those not familar with containers, Containers help modularize
services/applications, so afferent containers can hold different services e g a container can host the
front-end of your web app, and another container can host the back-end of your web app. Kubernetes
performs the automation, orchestration, management and deployment of your containers. GAE is a
platform for system developers, and as they say it best "Bring your code, we'l handie the rest". This
ensures customers that use GAE do not have to deal with the underlying hardware/misdeware but
can already have a pre-configured platform ready to go, all they need to do is provide the necessary
code required to run i. GAE automatically handles scaling to meet load and demand from users.
Heroku is a service that enables developers to build, run, and operate applications entirely in the
cloud Persistert disk and doud storage are typical cloud storage services offered via the Google Cloud
Platform.

Google has also rolled out some services specifically for various platforms such as connected home
which provides internet of Things services for home appliances, Android 05, an operating System for
hand held devices and WearOS, an operating system for wearable devices such as watches. Machines
in their data centers are commodity PCs which are organized into racks. Each machine costs about
$1000 and contains 2 terabytes dsk space, 16G RAM and stripped down Linux Kernel, Middieware and
Remote Procedure Calls are » common means of enforcing interoperability among different hardware
platforms that they

: Use the case study above to answer the following questions.

a) State 4 measures that Google has in place to ensure transparency and the type of Distributed
System transparency they are enforcing.(8)

b) From the case study above, give examples of services that are offered by Google based on the
models below:

I,Software as a Service

I, Platform as a Service

il, Infrastructure as a Service

c) Suggest 3 possible models that can be used to describe Google's system model. Justify each choice.
(6)

d) Use the passage above to explain the 4 premises of distributed systems.

e) State how Google has managed to handie the following in the implementation of their services;

i,Openness

ii, Heterozeneity

iii. Fault tolerance

iv. Concurrency

f) State how Google implements the following types of distributed systems:

I, distributed information system

Ii, pervasive systems

a) Measures that Google has in place to ensure transparency and the type of Distributed System
transparency they are enforcing:
1. **Service Level Agreements (SLAs):** Google provides clear SLAs to its customers, outlining the level
of service and availability they can expect. This ensures transparency regarding the performance metrics
and service guarantees. This relates to **Behavioral Transparency** in Distributed Systems, where the
behavior of the system is clear and predictable.

2. **Monitoring and Logging:** Google utilizes monitoring and logging tools to track the performance of
their services and provide insights into system operations. This transparency mechanism falls under
**Communication Transparency**, ensuring that system components can communicate effectively and
securely.

3. **Data Protection and Privacy Policies:** Google follows strict data protection and privacy policies,
detailing how user data is collected, stored, and used. By making these policies transparent to users,
Google enforces **Access Transparency**, ensuring that users have visibility into the handling of their
data.

4. **Security Audits and Compliance:** Google undergoes regular security audits and complies with
industry regulations to maintain a secure environment for their services. This reflects **Location
Transparency** in Distributed Systems, where users do not need to be aware of the physical location of
resources for access.

b) Examples of services offered by Google based on the models:

- **Software as a Service (SaaS):** Google offers services like GMail, Google Docs, and Heroku as
examples of Software as a Service where users can access software applications over the internet
without the need for local installation.

- **Platform as a Service (PaaS):** Google App Engine (GAE) is an example of Platform as a Service,
providing a platform for developers to build, deploy, and scale applications without managing the
underlying infrastructure.

- **Infrastructure as a Service (IaaS):** Google Compute Engine (GCE) and Google Kubernetes Engine
(GKE) are examples of Infrastructure as a Service, allowing users to create virtual machines and manage
containers in a flexible and scalable infrastructure.

c) Possible models to describe Google's system model:

1. **Cloud Computing Model:** Google's services align with a cloud computing model, where resources
are provided as services over the internet. This model allows for on-demand access to computing
resources, scalability, and flexibility.

2. **Microservices Architecture Model:** Google's use of containers, Kubernetes, and modular services
reflects a microservices architecture model. This approach enables flexibility, scalability, and
independent deployment of services.
3. **Platform-based Ecosystem Model:** Google's range of services, including software, platform, and
infrastructure offerings, form an ecosystem that caters to developers, businesses, and end-users. This
model emphasizes interconnectivity, integration, and ease of use across the Google ecosystem.

d) Premises of distributed systems explained in the context of Google:

1. **Concurrency:** Google's system supports multiple users and services simultaneously, handling
numerous requests and operations concurrently to ensure efficient use of computing resources.

2. **Scalability:** Google's services can scale to accommodate increasing demand, allowing for the
seamless expansion of resources and capabilities as needed.

3. **Fault Tolerance:** Google implements redundancy, backup systems, and failover mechanisms to
ensure service availability and uptime, even in the face of hardware failures or disruptions.

4. **Openness:** Google's services are accessible to a wide range of users and developers, with open
APIs, documentation, and collaboration opportunities promoting openness and interoperability.

e) How Google manages key aspects in the implementation of their services:

i. **Openness:** Google promotes openness through open APIs, developer tools, and documentation,
allowing users and developers to access and integrate with their services transparently.

ii. **Heterogeneity:** Google handles heterogeneity by supporting diverse platforms, devices, and
technologies within their ecosystem, ensuring compatibility and interoperability across different
systems.

iii. **Fault Tolerance:** Google employs redundancy, load balancing, and fault-tolerant architectures to
mitigate the impact of failures and ensure continuous service availability and reliability.

iv. **Concurrency:** Google manages concurrency by efficiently handling multiple requests and
operations simultaneously, utilizing parallel processing and distributed computing techniques to
maximize performance.

f) How Google implements types of distributed systems:

i. **Distributed Information System:** Google utilizes distributed databases, indexing, and caching
mechanisms to provide fast and reliable access to information across its services, ensuring data
availability and consistency.

ii. **Pervasive Systems:** Google's services extend beyond traditional computing devices, supporting
internet-connected home appliances, mobile devices, and wearable technology, creating a pervasive
ecosystem that integrates seamlessly into users' everyday lives.
SECTION B: AMPWER ANT 3 QUESTIONS I

IN THIS SECTION 100 MARICS)

2a)Explain what is meant by transparency in distribution

b) Define the following terms with respect to Distributed Systems and explain with examples where
possible how Hadoop enforces these features of distributed systems:

L. Distributed file storage

ii. Data locality

iii. Parallel processing of data

Iv. Fault tolerance

a) Transparency in distribution refers to the ability of a distributed system to hide the complexities of its
underlying architecture and provide a seamless user experience. There are various types of transparency
in distributed systems:

- **Access Transparency**: Users can access resources in a distributed system without needing to know
the physical location or distribution of those resources.

- **Location Transparency**: Users do not need to be aware of the physical location of resources in a
distributed system, allowing for seamless access regardless of where resources are located.

- **Migration Transparency**: The movement of resources between different locations in a distributed

system is transparent to users and applications.

- **Relocation Transparency**: Resources can be relocated within a distributed system without

affecting user access or application functionality.

- **Replication Transparency**: The replication of data or resources in a distributed system is

transparent to users, helping to enhance availability and reliability.

- **Failure Transparency**: Failures or disruptions in the system are handled transparently, ensuring
that users experience minimal impact.
- **Concurrency Transparency**: Users can perform concurrent operations in a distributed system
without needing to manage the complexities of synchronization and communication between multiple
processes.

b) Definitions and examples of how Hadoop enforces features of distributed systems:

i. **Distributed file storage**: Distributed file storage refers to the storage of files across multiple nodes
in a distributed system, enabling reliability, scalability, and fault tolerance. Hadoop enforces distributed
file storage through the Hadoop Distributed File System (HDFS), which stores large files across multiple
data nodes in a cluster. HDFS replicates data blocks across nodes to ensure fault tolerance and
availability. For example, when a file is uploaded to HDFS, it is split into blocks and distributed across
nodes in the cluster for parallel processing.

ii. **Data locality**: Data locality is the principle of processing data where it resides to minimize data
movement and improve performance in distributed systems. Hadoop enforces data locality by
scheduling tasks to run on nodes where data is stored, reducing network traffic and improving
processing efficiency. For example, when running a MapReduce job in Hadoop, tasks are scheduled to
process data blocks that are stored locally on the same node, maximizing data locality and minimizing
data transfer across the network.

iii. **Parallel processing of data**: Parallel processing of data involves splitting data into smaller chunks
and processing them simultaneously across multiple nodes in a distributed system to achieve faster
processing times. Hadoop enforces parallel processing through its MapReduce framework, which divides
data processing tasks into map and reduce phases that can run in parallel across nodes in a cluster. For
example, when executing a MapReduce job in Hadoop, data is processed in parallel across multiple
nodes, allowing for efficient computation and analysis.

iv. **Fault tolerance**: Fault tolerance in distributed systems refers to the system's ability to continue
functioning and serving users in the event of hardware failures, network interruptions, or other issues.
Hadoop enforces fault tolerance through data replication and job recovery mechanisms. For example, in
HDFS, data blocks are replicated across multiple nodes to ensure data availability in case of node
failures. Additionally, the MapReduce framework in Hadoop can restart failed tasks on other nodes to
complete processing jobs even in the presence of failures.

c) Explain how RPC can be used to ensure the following types of transparency:

1. Migration

2. Replication

3. Failure
Remote Procedure Call (RPC) is a communication protocol that allows a program to request a service
from a program located on a different computer in a network. RPC can be used to ensure different types
of transparency in a distributed system, including migration, replication, and failure transparency:

1. **Migration Transparency**:

Migration transparency refers to the ability of a distributed system to move resources or services from
one location to another without impacting users or applications. RPC can be used to achieve migration
transparency by abstracting the underlying details of the migration process from the users or
applications. When a service is migrated to a different server or location, RPC allows the client program
to continue making requests to the service without needing to know the new location. The RPC
framework handles the communication between the client and the migrated service transparently. This
ensures that users or applications can access the service seamlessly, regardless of its physical location.

2. **Replication Transparency**:

Replication transparency involves duplicating data or services across multiple nodes in a distributed
system to improve availability and reliability. RPC can be used to ensure replication transparency by
abstracting the replication process from users or applications. When a service is replicated on multiple
servers, RPC allows clients to access the service without needing to know which instance of the service
they are communicating with. The RPC framework manages the communication between the client and
the replicated services transparently, ensuring that requests are distributed evenly across the replicas.
This enables users or applications to access the service seamlessly, even in the presence of replication.

3. **Failure Transparency**:

Failure transparency refers to the ability of a distributed system to handle failures or disruptions without
impacting users or applications. RPC can be utilized to ensure failure transparency by providing
mechanisms for fault tolerance and recovery. In the event of a failure, such as a server crash or network
issue, RPC can handle the failure gracefully by rerouting requests to other available servers or nodes.
The RPC framework can automatically detect failures and redirect requests to alternative resources,
ensuring that users or applications experience minimal disruption. By abstracting the details of fault
tolerance from users or applications, RPC enables the system to maintain availability and reliability in
the face of failures.

Question 3

a) Describe 4 types of tramparency in distributed systems.

b) State the differences between the following:

i. Distributed systems versus parallel systems.

ii. Remote method invocation versus Remote procedure cal.

iii. Synchronous versus asynchronous communication

a)

1. **Access Transparency**: Users can access resources in a distributed system without needing to
know the physical location or distribution of those resources. This transparency allows users to interact
with resources as if they were all located locally, regardless of their actual location.

2. **Location Transparency**: Users do not need to be aware of the physical location of resources in a
distributed system, allowing for seamless access regardless of where resources are located. This
transparency enables the system to manage resource placement and movement without affecting user
interactions.

3. **Migration Transparency**: The movement of resources between different locations in a distributed

system is transparent to users and applications. This transparency allows resources to be relocated or
replicated without disrupting user access or application functionality.

4. **Replication Transparency**: The replication of data or resources in a distributed system is

transparent to users, helping to enhance availability and reliability. This transparency allows the system
to maintain multiple copies of data or services to improve fault tolerance and performance without
users needing to manage the replication process.

b)

i. **Distributed systems versus parallel systems**:

- Distributed systems involve multiple interconnected computers that communicate and coordinate with
each other to achieve a common goal. These systems typically span across different locations and are
designed to work together to solve complex problems. Parallel systems, on the other hand, involve
multiple processing units within a single computer or closely connected computers that work together
to execute tasks simultaneously. The main difference is that distributed systems focus on sharing
resources and collaborating over a network, while parallel systems focus on utilizing multiple processors
for simultaneous computation on a single machine or tightly coupled machines.

ii. **Remote method invocation versus Remote procedure call**:

- Remote method invocation (RMI) is a Java-based technology that allows a Java object to invoke
methods on an object running in another JVM, usually on a different machine. RMI is specific to Java and
enables objects to communicate and interact across a network. Remote procedure call (RPC), on the
 other hand, is a generic protocol that allows a program to execute procedures or functions on a remote
computer. RPC is language-independent and can be used to implement communication between
different programming languages or systems.

iii. **Synchronous versus asynchronous communication**:

- Synchronous communication is a communication method where the sender waits for a response from
the receiver before proceeding with other tasks. In this mode of communication, both the sender and
receiver are engaged in a real-time conversation. Asynchronous communication, on the other hand,
allows the sender to send a message to the receiver without waiting for an immediate response. The
sender can continue with other tasks, and the receiver processes the message at its own pace.
Asynchronous communication is typically used in scenarios where immediate responses are not
required or when there may be delays in message processing.

question 5

b) State 2 differences between persistent and transient messaging-

b) Wich the aid of a diagram, explain the role of middleware in a distributed system.

c) With the aid of examples, explain 3 benefits of middleware to programmers in a distributed system.

d) Explain why Interface Definition Languages are needed in distributed computing platforms.

1. backend servers. It also serves static content for browsing the online shop.
2. Backend servers: These servers handle the actual processing of requests,
such as retrieving product information, processing orders, and handling user
sessions.
3. Database server: This server stores product catalog data, user information,
and order details.
4. Load balancer: This component distributes incoming requests among multiple
backend servers to ensure load balancing and high availability.
5. Caching server: This server caches frequently accessed data to reduce the
load on the database server and improve performance.
b) Physical distribution of the software across the provider's machines:
1. Deploy frontend servers in multiple regions close to the clients to minimize
latency.
2. Deploy backend servers in locations with high demand to handle processing
of requests efficiently.
3. Utilize the provider's dynamic on-demand acquisition and relinquishment of
machines to scale up or down based on demand.
4. Use a global load balancer to distribute incoming traffic among frontend and
backend servers in different regions.
5. Implement a distributed caching mechanism to store frequently accessed data
close to the users.
c) Mechanisms and algorithms to handle request bursts:
1. Auto-scaling: Utilize auto-scaling capabilities provided by the cloud provider to
automatically add or remove machines based on traffic load.
2. Queueing: Implement a request queue to manage incoming requests during
burst periods and ensure fair processing.
3. Content delivery networks (CDNs): Cache static content on CDNs to reduce
server load during high traffic periods.
4. Dynamic load balancing: Use dynamic load balancing algorithms to distribute
traffic efficiently among available machines and maintain optimal
performance.
5. Resource utilization monitoring: Implement monitoring tools to track resource
usage and performance metrics to make data-driven decisions on scaling and
resource allocation.
d) Potential problems with the solutions:
1. Cost management: The automated scaling of resources can lead to
unexpected costs if not properly monitored and managed.
2. Data consistency: Distributing data across multiple locations can introduce
challenges related to data consistency and synchronization.
3. Security concerns: With a globally distributed system, maintaining security
standards and compliance across multiple regions can be complex.
4. Network latency: Distributing software across different regions can increase
network latency, affecting the overall performance of the system.
5. Dependency on cloud provider: The system's reliability is dependent on the
cloud provider's infrastructure and services, which can introduce risks in case
of provider downtime or issues.
CopySearch Web
Delete
 Delete
a) Fundamental models considered in the design of distributed systems:
1. Client-Server Model: In this model, clients request services or resources from
servers, which respond to these requests. It is a common model for
distributing processing across multiple machines and providing centralized
access to resources.
2. Peer-to-Peer Model: In this model, all nodes in the network can act as both
clients and servers, allowing for direct communication and resource sharing
between peers without the need for a central server. It is a decentralized
model that promotes scalability and fault tolerance.
3. Distributed Objects Model: In this model, objects are distributed across
different nodes in the network, and communication between objects is
achieved through method invocation. It allows for a more modular and flexible
design of distributed systems.
4. Message-Passing Model: In this model, communication between nodes is
achieved through message passing, where messages are sent and received
asynchronously. It provides a simple and efficient way for nodes to
communicate and exchange information.
b) Inter-process communication (IPC): Inter-process communication refers to
the mechanisms and techniques used for processes to communicate and
synchronize with each other in a distributed system. IPC is essential for
coordinating activities, sharing data, and achieving collaboration between
different processes or components within a distributed system. There are
several methods of IPC, including:
1. Shared Memory: In shared memory IPC, multiple processes can access a
common memory region that is shared among them. Processes can read from
and write to the shared memory segment, enabling fast communication and
data exchange.
2. Message Passing: Message passing IPC involves processes communicating
by sending and receiving messages. Messages can be of various types, such
as synchronous or asynchronous, and can contain data or commands. It
 provides a more flexible and secure way for processes to communicate
compared to shared memory.
3. Pipes: Pipes are a simple form of IPC that allows a unidirectional flow of data
between two processes. One process writes data to the pipe, and the other
process reads from it. Pipes are commonly used for communication between
a parent process and its child process.
4. Sockets: Sockets are endpoints for communication between processes over a
network. They can be used for IPC between processes running on the same
machine or on different machines. Sockets provide a versatile and efficient
way for processes to communicate over a network.
Overall, inter-process communication plays a crucial role in facilitating
coordination, data sharing, and collaboration between processes in a
distributed system, enabling them to work together towards achieving a
common goal.
CopySearch Web
Delete

Delete
a) In the context of moving towards web-based distributed computing, the
concern raised by the technical manager regarding the existing systems being
designed for CPU efficiency is well-founded. When transitioning to a
distributed system, there are several challenges that may arise, including:
1. Scalability: Web-based distributed systems often need to support a large
number of users and handle varying levels of traffic. The existing CPU-
efficient design may not be optimized for handling distributed workloads and
scaling resources dynamically.
2. Network Communication: Distributed systems rely on network communication
for inter-process communication and data exchange. The design
considerations for efficient CPU usage may not necessarily align with the
requirements for efficient network communication, which can impact overall
performance and latency.
3. Fault Tolerance: Distributed systems must be resilient to failures and errors
that can occur in a distributed environment. The existing system's focus on
CPU efficiency may not have addressed fault tolerance mechanisms such as
redundancy, replication, and error handling.
4. Data Consistency: Ensuring consistency and coherence of data across
distributed components can be challenging. The existing system's design may
not have considered the complexities of maintaining data consistency in a
distributed environment.
Therefore, it is important to reassess the design and architecture of the
existing system to adapt it to the requirements and challenges of web-based
distributed computing.
b) Synchronization becomes a more significant problem in a distributed
system with heterogeneous components due to the following reasons:
1. Clock Synchronization: Heterogeneous components may have different clock
speeds and time zones, making it challenging to synchronize events and
maintain a consistent notion of time across the system.
2. Communication Protocols: Different components may use varied
communication protocols and standards, leading to compatibility issues and
the need for protocol conversion and translation.
3. Data Representation: Heterogeneous components may store and process
data in different formats or data structures, requiring transformation and
mapping mechanisms for data interchange.
4. Resource Management: Synchronization of resources such as memory, disk
space, and processing power becomes complex in a heterogeneous
environment, as each component may have different capabilities and
requirements.
Overall, the diverse nature of heterogeneous components in a distributed
system introduces challenges in coordinating and synchronizing activities,
data, and resources efficiently.
c) Importance of caching in a distributed system to avoid inconsistent or stale
data:
1. Performance Improvement: Caching frequently accessed data locally can
reduce the latency of retrieving data from remote servers, improving overall
system performance and response times.
2. Data Consistency: Caching can help maintain data consistency by storing a
copy of the data closer to the requesting entity, reducing the likelihood of
accessing outdated or stale data from remote sources.
3. Scalability: Caching can help distribute the load on backend servers by
serving cached data locally, reducing the burden on central servers and
enabling horizontal scalability.
4. Resilience: In the event of network failures or downtime of remote servers,
cached data can still be accessed locally, ensuring continuous availability and
functionality of the system.
5. Cost-Effectiveness: Caching reduces the need for frequent access to remote
servers, resulting in cost savings on bandwidth and resource utilization.
Overall, caching plays a crucial role in improving performance, ensuring data
consistency, enhancing scalability, and providing resilience in a distributed
system, making it an essential component for optimizing system efficiency and
functionality.
CopySearch Web
Delete

Delete
a) Issues relating to deadlock in the proposed system: Deadlock can occur in
the booking process of allornothing.com if multiple users attempt to book the
same components at the same time, leading to a situation where each user
holds a resource required by the other to proceed. In this scenario, if one user
is waiting for a resource that is held by another user, a deadlock can occur,
preventing both users from completing their bookings.
b) Ways to avoid deadlock in the system:
1. Resource Allocation Strategy: Implement a resource allocation strategy that
prevents deadlock, such as utilizing a resource ordering policy where users
must acquire resources in a specific order to avoid circular waiting.
2. Timeout Mechanism: Implement a timeout mechanism that releases
resources held by a user if they do not complete the booking process within a
specified time limit, preventing resources from being unnecessarily locked.
3. Deadlock Detection: Implement a deadlock detection algorithm that
periodically checks for deadlock conditions and resolves them by rolling back
transactions or restarting the booking process.
4. System Design: Design the system in a way that minimizes the likelihood of
deadlock, such as limiting the number of resources that can be held
simultaneously or ensuring that resources are released promptly after use.
c) Preserving server state across web pages: To preserve server state across
multiple web pages during the booking process, the system can utilize
techniques such as:
1. Session Management: Utilize session variables to store user-specific
information and state across multiple web pages. Sessions can be maintained
using cookies, URL rewriting, or hidden form variables.
2. Hidden Form Fields: Store important data in hidden form fields within HTML
forms, allowing the server to retrieve and update the data as the user
navigates through different pages.
3. Server-side State Management: Use server-side databases or files to store
and retrieve session data, allowing the server to maintain state information
even when the user navigates away from the current page.
d) Ensuring system robustness in the face of users failing to complete the
booking process:
1. Transaction Rollback: Implement a transaction rollback mechanism that
cancels bookings and releases resources if the user fails to complete the
booking process or times out. This prevents incomplete bookings from
affecting system availability.
2. Error Handling: Provide clear error messages and prompts to guide users
through the booking process, informing them of any issues or incomplete
steps that need to be addressed.
3. Data Validation: Validate user input at each step of the booking process to
ensure that only valid and complete information is submitted, reducing the
likelihood of errors or incomplete bookings.
4. Booking Reservations: Reserve booked components for a limited time before
final confirmation and payment, allowing for temporary holds on resources to
prevent conflicts and ensure availability for other users.
CopySearch Web
Delete
 Delete
a) After the TC reboots following a crash, it must perform recovery actions to
ensure the integrity of the transaction. The following steps can be taken by the
TC after rebooting:
1. Checkpointing: The TC should have a mechanism to checkpoint the state of
the transaction before the crash. This checkpoint can help the TC resume the
transaction from where it left off.
2. Contact Participants: The TC should first contact the participants that it had
previously sent the COMMIT message to. It needs to ensure that the
participants have not made their updates permanent based on the COMMIT
message.
3. Resend Messages: If the participants have not committed their updates, the
TC needs to resend the COMMIT message to all participants to complete the
transaction. It should ensure all participants are in sync before proceeding
with the COMMIT phase.
4. Timeout Handling: The TC should handle timeout scenarios carefully to avoid
inconsistencies and ensure that all participants are properly coordinated.
5. Data Recovery: If necessary, the TC should handle data recovery and ensure
that any changes made by participants are consistent and durable.
b) It would not be appropriate for the client to find a new coordinator, TC2, and
ask TC2 to run two-phase commit again for the transaction because doing so
could lead to an incorrect outcome. Here is an example scenario to illustrate
why this approach is not advisable:
Scenario:
1. Initially, TC sends the PREPARE message to all participants, and all
participants respond with YES, indicating they are ready to commit.
2. However, before the TC can send the COMMIT message to all participants, it
crashes.
3. The client, unaware of the crash, decides to find a new coordinator, TC2, and
asks TC2 to run two-phase commit again for the transaction.
4. TC2 sends a new PREPARE message to all participants, and one of the
participants, Participant A, replies with NO this time, while other participants
respond with YES.
5. Based on the new responses, TC2 decides to abort the transaction and sends
ABORT messages to all participants.
6. Participant A, which had previously received a COMMIT message from the
first TC, has already made its updates permanent based on that message.
Receiving an ABORT message from TC2 now contradicts the previous
decision made during the first execution of two-phase commit.
This scenario demonstrates the potential inconsistency that can arise if a new
coordinator is introduced to run two-phase commit again without proper
coordination and awareness of the previous transaction state. It is crucial to
ensure that the transaction coordinator maintains consistency and integrity
throughout the entire transaction process.
CopySearch Web
Delete

Delete
a) Issues relating to security in the proposed system:
1. Data Privacy: When users provide personal and financial information to make
bookings on the website, there is a risk of this data being compromised if the
website's security measures are not robust. This could lead to identity theft,
fraud, or unauthorized access to sensitive information.
2. Payment Security: Since users will be making payments for their holiday
packages through the website, there is a concern about the security of
payment transactions. If the payment gateway is not secure, users' financial
information could be intercepted by malicious actors.
3. Data Integrity: There is a risk of data manipulation or corruption if the system
is not properly secured. Unauthorized parties could potentially modify
bookings, alter prices, or delete critical information.
b) Ways to avoid security issues:
1. Secure Socket Layer (SSL) Encryption: Implement SSL certificates to encrypt
data transmitted between users and the website, ensuring that information is
secure during transit.
2. Secure Payment Gateway: Use trusted and secure payment gateways that
comply with Payment Card Industry Data Security Standard (PCI DSS) to
protect users' financial information during transactions.
3. User Authentication: Implement secure login mechanisms, such as multi-
factor authentication, to verify the identity of users accessing the website and
prevent unauthorized access.
4. Data Encryption: Encrypt sensitive data stored in databases to prevent
unauthorized access and ensure data confidentiality.
5. Regular Security Audits: Conduct regular security audits and penetration
testing to identify and address vulnerabilities in the system proactively.
c) Making the system robust in the face of users failing to complete the
booking process:
1. Booking Timeout: Implement a booking timeout mechanism to release held
bookings if users fail to complete the transaction within a specified time
period, ensuring availability for other users.
2. Error Handling: Provide clear feedback and error messages to users during
the booking process to guide them in completing the transaction successfully.
3. Reservation System: Implement a reservation system that temporarily holds
bookings until final payment is made, allowing users some flexibility in
completing the booking process without losing their desired components.
4. Customer Support: Offer responsive customer support to assist users who
may encounter difficulties during the booking process, helping them to
complete their transactions successfully.
5. Transaction Logging: Maintain transaction logs to track the progress of
bookings and identify any incomplete transactions, allowing for follow-up and
resolution with the users.
CopySearch Web
Delete
 Delete
(a)
1. Concurrent execution of components:
 Implication: In a distributed system where components can execute
concurrently, engineers need to consider issues related to synchronization,
consistency, and concurrency control. This includes ensuring that shared
resources are accessed in a coordinated manner to prevent data corruption
and maintaining proper synchronization mechanisms to handle concurrent
requests effectively.
 Example: In a large-scale distributed system where multiple users are
accessing and updating a shared database concurrently, engineers need to
implement mechanisms like locks, transactions, and isolation levels to
maintain data consistency and avoid conflicts.
2. Independent failure modes:
 Implication: In a distributed system where components can fail independently,
engineers must design fault-tolerant mechanisms to handle failures gracefully
without compromising the system's availability and reliability. This includes
implementing redundancy, replication, and error detection and recovery
strategies to mitigate the impact of component failures.
 Example: In a widely distributed system where nodes can fail independently,
engineers can use techniques like replication and load balancing to ensure
that service availability is not affected if one or more components fail.
3. No global time:
 Implication: In a distributed system where there is no global time reference,
engineers need to address challenges related to event ordering,
synchronization, and ensuring consistency across distributed components.
This includes implementing protocols for clock synchronization, handling
causality constraints, and managing distributed transactions effectively.
 Example: In a large-scale distributed system where timestamps are used to
determine the order of events, engineers need to carefully design algorithms
and protocols to maintain causality and ensure that events are processed in
the correct order.
(b) (i) Role-Based Access Control (RBAC) is a security model that restricts
system access to authorized users based on their roles within an organization.
Each role is assigned specific permissions, and users are granted access
based on their role, rather than individual user identities.
 (ii) In a national healthcare system comprising multiple administration
domains, RBAC can be used to manage access control effectively. Principals,
such as healthcare professionals, can be assigned roles based on their
responsibilities and access requirements within different domains. For
example:
 A doctor working in a primary care practice may have a "Primary Care
Physician" role, granting access to patient records and treatment plans within
that domain.
 If the same doctor needs to work in a hospital or specialist clinic, they can be
assigned additional roles, such as "Hospital Physician" or "Specialist
Consultant," to access relevant information and systems in those domains.
(iii) To meet the requirement of capturing exclusions of principals and
relationships between them in an RBAC design for a national Electronic
Health Record (EHR) service, engineers can:
 Implement hierarchical roles: Define roles with varying levels of access
privileges and relationships between roles to capture complex authorization
requirements.
 Use constraints and role hierarchies: Establish exclusion rules and constraints
to restrict access to sensitive information based on relationships between
principals, ensuring that only authorized users can access certain data.
 Role activation and deactivation: Allow administrators to activate or deactivate
roles for principals based on their status, responsibilities, and affiliations within
different domains, ensuring that access rights are aligned with their current
roles and permissions.
CopySearch Web
Delete

Delete
a) Examples of applications in smartphones where message exchange is the
main purpose:
1. Messaging Apps: Messaging applications such as WhatsApp, iMessage, or
Facebook Messenger are specifically designed for users to exchange
messages with each other. The primary purpose of these apps is to facilitate
real-time communication through text, images, videos, and voice messages.
2. Email Apps: Email applications like Gmail, Outlook, or Apple Mail are used for
sending and receiving emails, where the main purpose is to exchange
messages electronically. These apps provide features for composing,
sending, receiving, and managing emails, making the message exchange
process efficient and user-friendly.
b) Diagram illustrating the essential difference between threaded and
distributed computations:
Threaded Computation:
 In threaded computation, multiple threads within a single process share the
same memory space and resources.
 Each thread can perform different tasks simultaneously, leveraging the
benefits of parallel processing.
 Threads communicate through shared memory and have a shared address
space.
 Diagram:

Copy Thread 1
| | |
V V V
Process
^ ^ ^
| | |
Thread 2

Distributed Computation:
 In distributed computation, multiple processes run on separate physical
machines or nodes connected through a network.
 Each process has its memory space and resources, and communication
between processes is achieved through message passing.
 Distributed systems can harness the power of multiple machines to perform
complex tasks in a decentralized manner.
 Diagram:

Copy Distributed Node 1 Distributed Node 2

| | | | | |
V V V V V V
Process Process Process Process

c) i) The scenario involving launching a word processor and a presentation

program on a modern desktop computer by a single user is likely to involve
concurrent computing. Concurrent computing refers to the execution of
multiple tasks simultaneously, which can happen on a single processor
through interleaved processing. In this scenario, both applications can run
concurrently, taking advantage of the operating system's scheduling
mechanism to switch between tasks and provide the illusion of parallel
execution.
ii) The scenario is not likely to involve distributed computing. Distributed
computing involves multiple independent computers (nodes) working together
to achieve a common goal. In the given scenario, the word processor and
presentation program are running on the same desktop computer, utilizing the
resources of a single machine rather than multiple distributed nodes.
CopySearch Web
Delete

Delete
The Byzantine Generals problem is a classic computer science problem that
deals with achieving consensus among a group of distributed entities when
some of these entities may be faulty or malicious. The problem is named after
the Byzantine Generals who faced a similar scenario where they needed to
reach an agreement on a coordinated attack or retreat, but some of the
generals could be traitors sending conflicting messages.
 In the context of distributed systems, the Byzantine Generals problem can be
stated as follows: a group of generals (nodes) must come to an agreement on
a common decision, such as attacking or retreating, despite the presence of
faulty or malicious nodes that may send conflicting or incorrect information.
The challenge is to reach consensus in the presence of Byzantine failures,
where faulty nodes may act arbitrarily and deceive other nodes.
Here are some notes on the Byzantine Generals problem and potential
solutions:
1. Challenges of the Byzantine Generals Problem:
 Byzantine failures: In a distributed system, nodes may fail in arbitrary ways,
including sending contradictory or incorrect messages.
 Lack of central authority: There is no central node that can be trusted to
provide correct information or make decisions for the entire system.
 Asynchronous communication: Nodes communicate over an unreliable
network where messages can be delayed, lost, or corrupted.
2. Solutions to the Byzantine Generals Problem:
 Byzantine Fault-Tolerant (BFT) algorithms: Byzantine fault-tolerant algorithms
are designed to achieve consensus in the presence of Byzantine failures.
These algorithms typically involve redundancy, replication, and cryptographic
techniques to tolerate Byzantine faults.
 Practical Byzantine Fault Tolerance (PBFT): PBFT is a popular Byzantine
fault-tolerant algorithm that uses a consensus protocol to reach an agreement
among nodes. PBFT ensures safety and liveness properties even in the
presence of Byzantine faults.
 Proof of Stake (PoS) and Proof of Work (PoW): Blockchain consensus
algorithms like PoS (used in Ethereum 2.0) and PoW (used in Bitcoin) provide
solutions to the Byzantine Generals problem by leveraging economic
incentives and cryptographic mechanisms to achieve consensus without a
trusted central authority.
 Hashgraph: Hashgraph is a distributed ledger technology that offers a
Byzantine fault-tolerant consensus algorithm based on gossip about gossip
and virtual voting.
 Practical applications: The Byzantine Generals problem has practical
applications in distributed databases, blockchain networks, decentralized
consensus protocols, and secure multi-party computations.
In conclusion, the Byzantine Generals problem highlights the challenges of
achieving consensus in distributed systems with faulty or malicious nodes. By
 employing Byzantine fault-tolerant algorithms, cryptography, and
decentralized consensus mechanisms, it is possible to address the Byzantine
Generals problem and ensure the integrity and reliability of distributed
systems in the face of Byzantine failures.
CopySearch Web
Delete

Delete
LLMs (Large Language Models) like the recently launched OpenAI's GPT-4
exemplify the true characteristics that define Distributed Systems by
showcasing massive scale, distributed computation, fault tolerance, and
decentralized decision-making. These characteristics present key challenges
that LLMs face, and innovative solutions are employed to address them
effectively.
1. Massive Scale: LLMs like GPT-4 operate on a massive scale, processing vast
amounts of data and performing complex computations. To handle this scale,
distributed systems are used to distribute the workload across multiple nodes
or servers. This allows for parallel processing, reducing the time required for
training and inference.
2. Distributed Computation: Distributed computation is essential for LLMs like
GPT-4, as training and running such large models require distributed
processing. By partitioning the model across different nodes and using parallel
computation, LLMs can handle the immense amount of data and calculations
efficiently.
3. Fault Tolerance: One of the key issues that LLMs face is the potential for
system failures or errors during computation. Distributed systems implement
fault tolerance mechanisms to ensure that the system can continue operating
even if individual nodes fail. Techniques such as replication, redundancy, and
checkpointing are used to recover from failures and maintain system
reliability.
4. Decentralized Decision-Making: LLMs like GPT-4 rely on decentralized
decision-making processes, where different nodes collaborate to reach a
 consensus on the next steps in the computation. Distributed systems utilize
consensus algorithms, such as the Byzantine fault-tolerant protocols, to
ensure that all nodes agree on the decisions being made, even in the
presence of faulty or malicious nodes.
5. Key Issues Faced by LLMs and Solutions: a) Data Processing Speed: LLMs
require high-speed data processing to train and run inference efficiently.
Distributed systems use parallel processing and distributed computing to
speed up data processing and reduce latency. b) Scalability: As the size of
LLMs grows, scalability becomes a challenge. Distributed systems can scale
horizontally by adding more nodes to the system, allowing LLMs to handle
larger models and datasets. c) Resource Management: LLMs require efficient
resource management to optimize performance. Distributed systems use
resource allocation algorithms and load balancing techniques to distribute
resources effectively and prevent bottlenecks. d) Communication Overhead:
Communication overhead in distributed systems can impact performance.
Techniques such as message passing optimization, data compression, and
efficient network protocols are used to minimize communication overhead in
LLMs.
In conclusion, LLMs like DeepAI's GPT-4 exemplify the characteristics of
Distributed Systems by leveraging distributed computation, fault tolerance,
scalability, and decentralized decision-making. By addressing key challenges
through innovative solutions, LLMs can overcome the complexities of large-
scale language modeling and deliver high-performance natural language
processing capabilities.
CopySearch Web
Delete

Delete
a) After the TC reboots following a crash before completing the two-phase
commit protocol for a transaction, it must perform the necessary recovery
steps to ensure the consistency and integrity of the transaction. The TC can
follow the following actions:
1. When the TC reboots, it should first check the status of the previous
transaction that failed to complete due to the crash. It needs to determine
whether any COMMIT messages were sent to the participants before the
crash.
2. If any COMMIT messages were sent but not received by all participants, the
TC should initiate a recovery procedure. It can communicate with the
participants to check if they received the COMMIT message and acted upon
it.
3. If all participants have not received the COMMIT message or have not
committed their parts of the transaction, the TC should abort the transaction
by sending an ABORT message to all participants. This will ensure that the
transaction is rolled back and the system returns to a consistent state.
b) It would not be acceptable for the client to find a new coordinator, TC2, and
ask TC2 to run the two-phase commit again for the transaction because this
could lead to an incorrect outcome due to the following scenario: Imagine a
situation where the original TC had sent a COMMIT message to one of the
participants in the transaction before crashing. The participant that received
the COMMIT message may have already committed its part of the transaction
and made the updates permanent in the database. If the client were to
engage a new coordinator, TC2, and restart the two-phase commit process
from the beginning, including sending PREPARE messages again, the
previously committed participant may receive a PREPARE message from
TC2. Since the participant has already committed its part of the transaction
and made the updates permanent, it may reply with a NO instead of YES in
response to the PREPARE message. This would result in an inconsistent
state where some participants have committed the transaction, while others
have not. This violates the atomicity property of transactions, where all
participants should either commit or abort the transaction together to maintain
data consistency. Therefore, it is crucial for the TC to handle the recovery
process correctly and maintain the integrity of the transaction to prevent such
inconsistencies in the database.
CopySearch Web
Delete
 Delete
a) Logical Organization of the Software:
1. Frontend Servers: Handle client requests, serve web pages, and process user
interactions.
2. Backend Servers: Manage inventory, process orders, and handle databases.
3. Load Balancer: Distribute incoming traffic across multiple frontend servers for
load balancing and high availability.
4. Caching Mechanism: Cache frequently accessed data to reduce database
load.
5. Queueing System: Queue requests during bursts to prevent overwhelming the
system.
6. Monitoring System: Monitor system performance and resource usage.
b) Physical Distribution of the Software:
1. Deploy frontend servers in various regions to reduce latency for clients.
2. Backend servers can be distributed for redundancy and data locality.
3. Use cloud provider's regions to ensure availability and scalability.
4. Deploy load balancers in front of frontend servers to distribute traffic
efficiently.
c) Mechanisms and Algorithms for Handling Request Bursts:
1. Auto-Scaling: Automatically scale resources based on predefined thresholds
or demand spikes.
2. Horizontal Scaling: Add more instances of software across different provider
machines during bursts.
3. CDN (Content Delivery Network): Serve static content from edge servers
closer to clients.
4. Caching Strategies: Implement caching mechanisms to reduce database load
and serve static content quickly.
5. Request Throttling: Limit the number of requests accepted during bursts to
prevent system overload.
d) Potential Problems with the Solutions:
1. Cost Management: Auto-scaling and dynamic resource allocation could lead
to unexpected cost increases if not properly managed.
2. Data Consistency: Maintaining data consistency across distributed systems
can be challenging, especially during bursts.
3. Performance Degradation: Load balancing algorithms may not always
distribute traffic efficiently, leading to performance issues.
4. Security Concerns: Distributing software across multiple machines increases
the attack surface, requiring robust security measures.
5. Resource Wastage: Instances may be running unnecessarily during off-peak
hours, leading to resource wastage and increased costs.
CopySearch Web
Delete

Delete
a) Fundamental Models in the Design of Distributed Systems:
1. Client-Server Model: In this model, the system is divided into clients that
request services or resources and servers that provide those services or
resources. Clients initiate requests, and servers respond to them, enabling a
clear separation of concerns and scalability in distributed systems.
2. Peer-to-Peer Model: In a peer-to-peer model, all nodes in the system have
equal responsibilities and capabilities. Nodes can act as both clients and
servers, sharing resources and information directly with one another without
the need for a central server.
3. Hierarchical Model: The hierarchical model organizes nodes in a tree-like
structure, with higher-level nodes managing and coordinating lower-level
 nodes. This model offers scalability, fault tolerance, and easier management
of the distributed system.
4. Distributed Objects Model: In this model, distributed components
communicate by invoking methods on remote objects. Objects encapsulate
both data and behavior, allowing for easy distribution and communication
between different components in the system.
5. Event-Driven Model: In an event-driven model, components in the distributed
system communicate through events and messages. Components can publish
events, subscribe to events, and react to events, enabling loosely coupled and
scalable systems.
b) Inter-Process Communication (IPC): Inter-process communication refers to
the mechanisms and techniques used by processes running on different
nodes or machines to communicate with each other in a distributed system.
IPC allows processes to exchange data, synchronize activities, and
coordinate their operations. Some common methods of IPC include:
1. Message Passing: Processes communicate by sending messages to each
other. Messages can be synchronous or asynchronous and can contain data
or control information.
2. Remote Procedure Call (RPC): RPC allows a process to execute a procedure
or method on a remote machine as if it were a local procedure call. The calling
process sends a request to the remote process, which executes the
procedure and returns the result.
3. Shared Memory: Processes can communicate by sharing a common memory
segment. Changes made by one process to the shared memory are visible to
other processes, enabling fast and efficient communication.
4. Sockets: Processes can communicate over a network using socket
programming. Sockets provide an interface for processes to send and receive
data over a network, enabling communication between processes running on
different machines.
Using IPC mechanisms, processes in a distributed system can communicate,
coordinate their actions, share resources, and work together to achieve
common goals. Effective IPC implementation is essential for designing
reliable, responsive, and scalable distributed systems.
CopySearch Web
Delete
 Delete
a) Comment on Concerns Regarding CPU Efficiency in Web-Based
Distributed Computing: It is understandable that the technical manager is
concerned about the transition to web-based distributed computing from
systems that were originally designed to be CPU-efficient. Web-based
distributed computing involves communication between different components
over a network, which can introduce additional overhead and resource
utilization compared to traditional CPU-bound applications. The increased
network latency, data serialization/deserialization, and coordination among
distributed components can potentially impact CPU efficiency.
However, the concerns are not unfounded. In a distributed system, the focus
shifts from optimizing CPU performance to optimizing network communication,
data transfer, and overall system scalability. While CPU efficiency remains
important, other factors such as latency, bandwidth, and fault tolerance
become equally significant in distributed systems. As a software engineer, it is
essential to strike a balance between CPU efficiency and overall system
performance to ensure the successful transition to web-based distributed
computing.
b) Synchronization Challenges in Heterogeneous Distributed Systems: In a
distributed system where components are very heterogeneous,
synchronization becomes a more significant problem due to the following
reasons:
1. Different Architectures: Heterogeneous components may run on different
platforms or have varying hardware configurations, making it challenging to
synchronize data and operations effectively.
2. Diverse Communication Protocols: Heterogeneous systems may use different
communication protocols or data formats, complicating the exchange of
information and coordination between components.
3. Varying Processing Speeds: Components with different processing speeds or
capabilities may lead to synchronization issues, such as race conditions or
data inconsistency.
4. Time Synchronization: Ensuring accurate time synchronization across
heterogeneous components is crucial for coordinating events and maintaining
system consistency.
To address synchronization challenges in heterogeneous distributed systems,
software engineers must design robust synchronization mechanisms, use
standard communication protocols, implement data serialization techniques,
and ensure proper error handling and conflict resolution strategies.
c) Importance of Caching in Distributed Systems: Caching plays a crucial role
in improving performance, reducing latency, and avoiding inconsistent or stale
data in distributed systems by:
1. Reducing Network Traffic: Caching frequently accessed data or results locally
at each node helps reduce the need to fetch data over the network, thereby
decreasing network traffic and latency.
2. Enhancing Scalability: Caching can help distribute the load evenly by serving
cached data quickly without overloading the backend servers, leading to
better scalability and resource utilization.
3. Improving Response Time: Cached data can be served quickly to clients,
improving response times and overall system performance.
4. Mitigating Data Inconsistency: By maintaining a consistent cache invalidation
strategy and updating cached data regularly, caching helps avoid serving
stale or outdated information to clients.
5. Enhancing Availability: Caching can improve the availability of data by serving
cached copies even in the event of network failures or server downtime,
ensuring uninterrupted service for users.
Overall, caching in distributed systems is essential for optimizing
performance, reducing resource consumption, improving user experience, and
maintaining data consistency across the system. Proper caching strategies
are vital for maximizing the benefits of distributed computing while minimizing
potential drawbacks related to data inconsistency or staleness.
CopySearch Web
Delete
 Delete
a) Peer-to-Peer Architectural Model for a Named Distributed System: One
example of a peer-to-peer architectural model for a distributed system is the
BitTorrent protocol, commonly used for file sharing over the internet. In a
BitTorrent network, users (peers) share files directly with each other without
the need for a central server. The key components of the BitTorrent peer-to-
peer architectural model include:
 Tracker: A centralized server that helps peers discover each other and
coordinate the sharing of files.
 Peers: Individual users sharing and downloading files in the network.
 Torrent File: Contains metadata about the files being shared, including the
tracker's information and file details.
 Pieces: Files are split into smaller pieces that are shared among peers to
improve download efficiency.
 Swarm: The group of peers sharing a specific file is known as a swarm.
In the BitTorrent peer-to-peer model, each peer can act as both a client and a
server, downloading and uploading pieces of the file simultaneously. Peers
connect to the tracker to find other peers, exchange pieces of the file, and
maintain a healthy swarm for efficient file sharing.
b) Asymmetric Cryptography Technique in Distributed Systems Security:
Asymmetric cryptography, also known as public-key cryptography, uses key
pairs (public and private keys) to secure communication and authenticate
entities in a distributed system. The public key is shared openly, while the
private key is kept secret. Asymmetric cryptography can be used in supporting
security in distributed systems in the following ways:
 Encryption: Public keys can be used to encrypt data that can only be
decrypted using the corresponding private key. This ensures confidentiality
and privacy in data transmission.
 Digital Signatures: Private keys can be used to create digital signatures that
verify the authenticity and integrity of data or messages, allowing recipients to
verify the sender's identity and detect tampering.
 Key Exchange: Public-key cryptography facilitates secure key exchange
protocols, such as Diffie-Hellman key exchange, to establish shared secret
keys for secure communication.
 Secure Communication Channels: Asymmetric encryption can help establish
secure channels between distributed system components, protecting data in
transit from unauthorized access or modification.
By leveraging asymmetric cryptography techniques, distributed systems can
achieve secure authentication, confidentiality, data integrity, and key
exchange mechanisms to mitigate security threats and vulnerabilities.
c) Synchronization in Distributed Systems: Synchronization in distributed
systems refers to the coordination of activities, data access, and
communication among multiple nodes or components to ensure consistency
and coherence. Key aspects of synchronization in distributed systems include:
 Mutual Exclusion: Ensuring that only one node can access a shared resource
at a time to prevent conflicts and data corruption.
 Concurrent Access: Allowing multiple nodes to access data concurrently while
maintaining consistency and avoiding race conditions.
 Consistency Models: Defining rules and protocols for maintaining data
consistency across distributed nodes, such as eventual consistency, strong
consistency, or causal consistency.
 Coordination Protocols: Implementing synchronization mechanisms, such as
distributed locks, timestamps, version vectors, or distributed transactions, to
manage concurrent access and maintain system integrity.
Synchronization in distributed systems is essential for ensuring data reliability,
transaction integrity, and system correctness across geographically dispersed
nodes. Effective synchronization mechanisms are crucial for maintaining
system coherence, preventing data inconsistencies, and supporting reliable
communication and collaboration in distributed environments.
CopySearch Web
Delete
 Delete
a) Three Benefits of Using Virtual Machines in Cloud Computing:
1. Cost Efficiency: By assigning virtual machines (VMs) instead of physical
machines, cloud computing services like Amazon's EC2 can optimize
resource utilization and reduce operational costs. VMs allow for efficient
allocation of computing resources based on demand, enabling users to pay
for only the resources they use. This pay-as-you-go model helps businesses
save money by avoiding upfront infrastructure investments and scaling
resources as needed, leading to cost efficiency and scalability.
2. Flexibility and Scalability: Virtual machines offer flexibility in terms of resource
allocation, configuration, and deployment. Users can easily create, modify,
and resize VM instances based on their requirements without the limitations of
physical hardware. Cloud providers like Amazon can quickly provision and
scale VMs to meet varying workloads, ensuring high availability and
performance. This flexibility and scalability enable businesses to adapt to
changing demands and optimize resource utilization efficiently.
3. Resource Isolation and Management: Virtual machines provide isolation
between different users and applications running on the same physical
infrastructure, enhancing security and performance. Amazon's EC2 can
allocate VMs to users with dedicated computing resources and customized
configurations, ensuring the isolation of workloads and data. VM management
tools enable efficient monitoring, provisioning, and orchestration of VM
instances, simplifying resource management and optimizing workload
distribution in a cloud environment.
b) Benefits of Using Cloud Computing for a Supermarket Chain:
1. Scalability: Cloud computing allows a supermarket chain to scale its IT
infrastructure according to seasonal demand, promotions, or new store
openings. The supermarket chain can easily provision additional resources,
such as storage and computing power, to handle increased website traffic,
inventory management, customer data analysis, and online ordering services
during peak times.
2. Cost Efficiency: Utilizing cloud services eliminates the need for the
supermarket chain to invest in and maintain on-premises hardware and
infrastructure. Cloud computing offers a pay-as-you-go model, enabling the
supermarket chain to pay for only the resources they consume, reducing
operational costs, and avoiding unnecessary capital expenditures on IT
infrastructure.
3. Data Security and Backup: Cloud computing provides robust data security
measures, such as encryption, access control, and data backup, ensuring the
protection and integrity of sensitive customer information and business data.
The supermarket chain can leverage cloud-based data storage and backup
services to prevent data loss, mitigate risks of hardware failures, and enhance
disaster recovery capabilities.
c) Differences Between Grid Computing and Cluster Computing:
1. Architecture: Grid computing involves coordinating resources from multiple
distributed and independent sources to solve large-scale computational
problems, focusing on resource sharing and collaboration. In contrast, cluster
computing connects multiple computers or servers within a local network to
work on a single task, emphasizing parallel processing and high-performance
computing within a centralized system.
2. Resource Ownership: In grid computing, resources are owned by different
organizations or individuals, and cooperation is based on a shared
infrastructure for resource utilization. Cluster computing typically involves a
single organization or entity owning and managing the cluster of machines,
leading to centralized control and resource allocation within the cluster.
3. Task Allocation: Grid computing assigns tasks to available resources
dynamically based on resource availability and capabilities across the grid
network, enabling efficient load balancing and optimization. Cluster computing
allocates tasks to specific nodes within the cluster based on predefined
configurations, ensuring dedicated resources and performance isolation for
each task or application running on the cluster.
CopySearch Web
Delete
 Delete
a) Design of a Distributed System Implementing Edge Computing Principles:
Edge computing involves moving computation and data processing closer to
the edge of the network, near where data is generated, rather than relying
solely on centralized cloud servers. A distributed system designed with edge
computing principles would involve deploying computing resources (such as
servers, storage, and processing units) at the network edge to enable faster
data processing, reduce latency, and enhance overall system efficiency. The
design of such a distributed system could include the following components:
1. Edge Devices: Devices such as sensors, cameras, and IoT devices that
generate data at the network edge.
2. Edge Servers: Servers deployed at the edge of the network to process data
locally and handle computing tasks closer to where data is generated.
3. Edge Gateways: Devices that connect edge devices to the edge servers and
facilitate data transfer and communication between devices and servers.
4. Distributed Data Storage: Storage systems distributed at the edge to store
and manage data closer to where it is generated, reducing latency and
improving data access.
5. Edge Computing Software: Software applications and services designed to
run on edge servers to process data, run analytics, and enable real-time
decision-making at the network edge.
6. Communication Protocols: Protocols and standards for efficient
communication between edge devices, gateways, servers, and cloud
resources, ensuring seamless data flow and system integration.
The design of a distributed system implementing edge computing principles
aims to enhance performance, reduce bandwidth usage, improve scalability,
and support real-time applications by leveraging distributed resources at the
network edge.
b) Recent Developments in Hardware and Communication Technologies
Contributing to IoT:
1. Miniaturization and Low Power Consumption: Recent advancements in
hardware technology have led to the development of small, low-power devices
with enhanced computing capabilities. Miniaturized sensors, processors, and
communication modules enable the creation of IoT devices that can operate
efficiently on battery power for extended periods, supporting the proliferation
of IoT applications in various industries.
2. Wireless Communication Standards: The evolution of wireless communication
technologies, such as Bluetooth Low Energy (BLE), Zigbee, LoRa, and 5G
 networks, has enabled seamless connectivity and data exchange between IoT
devices and networks. These communication standards support reliable data
transmission, low latency, and improved network coverage, facilitating the
growth of IoT ecosystems and enabling diverse IoT solutions.
3. Edge Computing Infrastructure: The rise of edge computing architectures and
edge devices with enhanced processing capabilities has contributed to the
development of IoT systems that can perform data processing and analytics
locally, reducing latency and enhancing real-time decision-making. Edge
computing infrastructure complements cloud services by offloading
computation tasks to edge servers, supporting IoT applications that require
low latency, high availability, and efficient resource utilization.
4. Security and Privacy Measures: Recent advancements in hardware
encryption, secure boot mechanisms, and firmware updates have
strengthened the security and privacy features of IoT devices and networks.
Hardware-based security solutions, such as Trusted Platform Modules (TPMs)
and Secure Element chips, protect IoT devices from cyber threats,
unauthorized access, and data breaches, ensuring the integrity and
confidentiality of IoT data and communications.
Overall, recent developments in hardware and communication technologies
have accelerated the growth of IoT ecosystems, driving innovation, improving
connectivity, and enabling diverse IoT applications across industries such as
healthcare, smart homes, transportation, and industrial automation.
CopySearch Web
Delete

Delete
I'm unable to view Figure 4 as it was not provided in the question. However, I
can explain how MapReduce works in the context of processing a bag of
words.
MapReduce is a programming model and processing framework used to
process large datasets in a distributed computing environment. The basic idea
behind MapReduce is to break down a complex computation task into smaller,
 independent tasks that can be processed in parallel across multiple nodes in a
cluster. The MapReduce framework consists of two main phases: the Map
phase and the Reduce phase.
1. Map Phase:
 In the Map phase, the input data (in this case, the bag of words) is divided into
key-value pairs. Each key-value pair represents a unit of data that will be
processed independently.
 The Map function processes each key-value pair and generates intermediate
key-value pairs as output. For a bag of words, the Map function can tokenize
the input text into individual words and assign a count of 1 to each word as the
value.
 The intermediate key-value pairs are then shuffled and sorted based on their
keys to group together values associated with the same key.
2. Shuffle and Sort:
 The shuffle and sort phase reorganizes the intermediate key-value pairs
generated by the Map function. All values with the same key from different
nodes are grouped together.
 This phase ensures that all values corresponding to the same key are
processed together during the Reduce phase.
3. Reduce Phase:
 In the Reduce phase, each unique key and its corresponding list of values are
passed to the Reduce function. The Reduce function aggregates and
processes these values to produce the final output.
 For a bag of words, the Reduce function can sum up the word counts for each
unique word to generate a final word frequency count.
By utilizing the MapReduce framework, the processing of a bag of words
dataset can be distributed across multiple nodes in a cluster, allowing for
parallel computation and efficient processing of large volumes of data.
MapReduce simplifies the task of processing and analyzing big data by
breaking it down into smaller, manageable chunks that can be processed in
parallel, leading to faster computation and scalability.
CopySearch Web
Delete
 Delete
a) Benefits of Middleware to Distributed Systems Programmers:
i. Simplified Communication: Middleware provides a layer of abstraction that
simplifies communication between different components in a distributed
system. It hides the complexities of low-level network protocols and provides
high-level APIs for developers to easily exchange messages and data.
Example: Consider a distributed system where multiple microservices need to
communicate with each other. By using middleware like Apache Kafka or
RabbitMQ, programmers can publish and subscribe to messages using topics
or queues without worrying about the underlying network details. This
simplifies the development process and makes it easier to establish
communication channels between different components.
ii. Scalability and Load Balancing: Middleware often includes features for
managing load balancing and scaling distributed applications. It can distribute
incoming requests across multiple servers or instances to ensure optimal
performance and resource utilization.
Example: In a distributed e-commerce system, middleware like Nginx or
HAProxy can be used to balance the load between multiple web servers
hosting the online store. It can direct user requests to the least busy server,
distributing the workload evenly and preventing any single server from
becoming overwhelmed. This helps in achieving high availability and
scalability for the system.
b) System Availability Calculations: i. If the system is designed so that the
system can be operational if any one of the four servers is operational:
 The overall system availability can be calculated using the formula for multiple
redundant components in parallel: Total Availability = 1 - ((1 - individual
component availability)^number of components) Total Availability = 1 - ((1 -
0.9)^4) = 1 - (0.1^4) = 1 - 0.0001 = 0.9999 or 99.99%
ii. If all four servers have to be available for the entire system to be available:
 In this case, the overall system availability would be equal to the availability of
the least available component, which is 90% in this scenario.
c) Advantages of Peer-to-Peer Systems:
1. Decentralization: Peer-to-peer systems do not rely on central servers or
infrastructure, allowing for greater resilience and fault tolerance. If one peer
fails, the network can still operate.
2. Scalability: Peer-to-peer systems can scale efficiently as each peer
contributes resources to the network, allowing for the expansion of services
and data storage without relying on dedicated servers.
3. Anonymity and Privacy: Peer-to-peer systems offer increased privacy and
anonymity as peers communicate directly with each other without the need for
centralized intermediaries, protecting user data.
4. Reduced Costs: Peer-to-peer systems can lower costs by distributing the
workload and resource requirements among peers, reducing the need for
expensive server infrastructure and maintenance.
d) Difference between Distributed Operating System and Network Operating
System:
 A Distributed Operating System (DOS) is an operating system that runs on a
network of computers and manages resources across multiple nodes in the
network. It provides a unified interface for users to access resources
distributed across the network.
 A Network Operating System (NOS) is an operating system specifically
designed for managing network resources and providing network services,
such as file sharing, printer access, and user authentication. It focuses on
communication and coordination between devices on a network.
In summary, a DOS focuses on managing resources and coordinating tasks
across a distributed system of computers, while a NOS is focused on
providing network services and communication capabilities within a local area
network (LAN) or wide area network (WAN).
CopySearch Web
Delete
 Delete
a) Comparison of Cloud Computing with Traditional Client-Server Computing:
 In traditional client-server computing, data and applications are stored and run
on dedicated servers within an organization's premises, requiring physical
maintenance and IT resources. In contrast, cloud computing relies on remote
servers hosted by third-party providers, allowing organizations to access
services and resources over the internet.
 Traditional client-server models often require significant upfront costs for
hardware and software procurement, while cloud computing offers a pay-as-
you-go model, where organizations only pay for the resources they consume.
 Cloud computing provides scalability and flexibility, allowing organizations to
easily scale resources up or down based on demand, while traditional client-
server setups may require additional hardware or infrastructure upgrades to
accommodate growth.
b) Novelty of Cloud Computing:
 One novel aspect of cloud computing is the concept of virtualization, where
physical hardware resources are abstracted and pooled together, allowing for
efficient utilization and allocation of resources.
 Another novel aspect is the self-service provisioning model, where users can
easily request and deploy resources without needing direct intervention from
IT administrators. This on-demand access to resources promotes agility and
flexibility within organizations.
c) On-Demand Access Services Enabled by Cloud Computing: i. IaaS
(Infrastructure as a Service): With IaaS, organizations can access virtualized
computing resources such as virtual machines, storage, and networking on a
pay-as-you-go basis. Users can quickly provision and scale these resources
based on their requirements without investing in physical hardware. ii. SaaS
(Software as a Service): SaaS allows organizations to access and use
software applications hosted on the cloud. Users can access these
applications through a web browser without the need for installation or
 maintenance. Examples of SaaS include Office 365, Salesforce, and Google
Workspace.
d) Justification for Adoption of Cloud Computing for an Organization:
 Example Organization: XYZ Company is a medium-sized manufacturing
company with multiple locations and a growing need for digital collaboration
and data storage.
 Justification: XYZ Company should adopt cloud computing for their
organization to leverage the benefits of scalability, cost-efficiency, and
improved collaboration.
 By migrating their data and applications to the cloud, XYZ Company can
easily scale their resources based on demand and reduce upfront costs
associated with maintaining physical servers.
 Cloud-based collaboration tools and services can enhance communication
and productivity among employees in different locations, promoting better
teamwork and efficiency.
 Additionally, cloud-based data storage and backup solutions can provide
increased security, reliability, and accessibility to critical business information,
ensuring business continuity and disaster recovery capabilities for XYZ
Company.
CopySearch Web
Delete

Delete
a) Distributed Database Protocols Comparison:
i) Two Phase Commit protocol (2PC):
 In the scenario provided, if the coordinator crashes immediately after sending
the "DoCommit" message to server A but before sending a message to server
B, server A will commit its changes to its local database since it received the
"DoCommit" message. However, server B will not commit its changes as it did
not receive any instruction from the coordinator.
 ii) Paxos Consensus Protocol:
 In a scenario where a proposer waits for less than a majority of acceptors to
answer OK to a Prepare or an Accept message before proceeding to the next
steps, there is a risk of violating Paxos' service guarantees. If the proposer
proceeds before receiving responses from a majority, it may lead to
inconsistencies and potential conflicts in the consensus process, impacting
the correctness of the consensus outcome.
For Paxos, calculate the number of messages exchanged to update the
balance of a single account with 100 servers:
 In Paxos, the number of messages exchanged to update the balance of a
single account would involve multiple rounds of Prepare, Accept, and Commit
messages exchanged between the proposer and acceptors. The specific
calculation of the exact number of messages exchanged would depend on the
implementation details and the number of rounds required to achieve
consensus.
b) Preventing Unnecessary View Changes in Consensus:
 Unnecessary view changes, also known as leader changes in distributed
systems, can hinder consensus progress by introducing delays and
disruptions in the decision-making process.
 One mechanism to prevent unnecessary view changes is to implement a
heartbeat mechanism where the leader periodically sends heartbeat
messages to all nodes in the system to indicate its liveness and availability. If
a node does not receive a heartbeat from the leader within a certain interval, it
can trigger a view change process to elect a new leader.
 Additionally, introducing a threshold or quorum-based approach can help
prevent premature view changes by ensuring that a certain number of nodes
acknowledge the need for a view change before proceeding with the transition
to a new view.
 By carefully monitoring the leader's activity, implementing timeout
mechanisms, and considering a consensus-based approach for initiating view
changes, unnecessary view changes can be minimized, ensuring the stability
and progress of the consensus protocol in practice.
CopySearch Web
Delete
Delete
a) Challenges in Distributed Systems and Addressing Them:
i) Scalability: As the number of nodes and users in a distributed system grows,
scalability becomes a critical challenge. To address this, techniques such as
load balancing, sharding, and vertical/horizontal scaling can be implemented.
Load balancing distributes incoming requests among multiple servers to
optimize resource usage. Sharding involves partitioning the data across
multiple servers to distribute the workload evenly. Vertical scaling refers to
increasing the resources of individual nodes, while horizontal scaling involves
adding more nodes to the system.
ii) Fault Tolerance: Distributed systems are prone to failures, leading to data
inconsistencies and service disruptions. To ensure fault tolerance,
mechanisms such as replication, redundancy, and fault detection can be
employed. Replication involves creating copies of data across multiple nodes
to ensure data availability in case of failures. Redundancy involves having
backup systems or components to take over in case of failures. Fault
detection mechanisms monitor the system for failures and trigger recovery
actions automatically.
b) Virtual Organization:
A virtual organization is a network of geographically dispersed individuals or
groups linked together through information and communication technologies
to achieve common goals. These organizations operate without physical
boundaries, utilizing virtual teams and platforms for collaboration and
communication. Implementing virtual organizations involves leveraging tools
such as virtual meeting software, project management platforms, and cloud-
based collaboration tools. Additionally, establishing clear communication
channels, setting goals and expectations, and fostering a culture of trust and
accountability are essential elements of virtual organizations.
c) Features of Distributed Pervasive Systems in Railway Station Scenario:
 Context Awareness: The distributed pervasive system at the railway station
can utilize sensors and location-based services to gather information about
the user's surroundings and context. This information can include the user's
location, the time of day, available services, and amenities at the station.
 Seamless Connectivity: The mobile device's wireless networking capabilities
allow the user to access information about local services and amenities
without the need to explicitly enter any details. The distributed pervasive
system enables seamless connectivity and information retrieval based on the
user's context.
 Ubiquitous Access: The distributed pervasive system ensures that the user
can access relevant information about the railway station's services and
amenities from anywhere within the station premises. This ubiquitous access
enhances the user experience and provides convenience.
 Real-time Updates: The distributed system can provide real-time updates and
notifications to the user about train schedules, platform changes, delays, and
other relevant information. This ensures timely and accurate information
delivery to enhance the user's overall experience at the railway station.
CopySearch Web
Delete

Delete
a) Nakamoto Consensus:
The Nakamoto Consensus, also known as Proof of Work (PoW), is a
consensus mechanism used in blockchain technology to validate transactions
and secure the network. In the Nakamoto Consensus, miners compete to
solve complex mathematical puzzles in order to add new blocks to the
blockchain. The first miner to solve the puzzle broadcasts the solution to the
network, and if the solution is verified by other nodes, the new block is added
to the blockchain. This process requires significant computational power and
energy to deter malicious actors from manipulating the blockchain.
b) Improvements in Conflux Decentralized Blockchain System:
Conflux is a decentralized blockchain system that has made several
improvements to address limitations of the Nakamoto Consensus, including:
i) Scalability: Conflux uses a novel parallelized architecture that allows for
faster transaction processing and higher throughput compared to traditional
blockchain systems. By running consensus algorithms in parallel, Conflux can
achieve higher scalability without sacrificing security.
ii) Low Latency: Conflux utilizes a Tree Graph structure that enables faster
block propagation and confirmation times. This reduces latency in transaction
processing and improves the overall efficiency of the blockchain network.
iii) Fairness: Conflux incorporates a decentralized governance model that
ensures fairness in decision-making processes within the network. This allows
for community-driven governance and prevents centralization of power.
iv) Security: Conflux integrates advanced cryptographic techniques and
security measures to protect against attacks and ensure the integrity of the
blockchain. By incorporating robust security mechanisms, Conflux enhances
the overall security of the decentralized system.
c) Designing a Blockchain System to Solve a Real-World Problem:
To use blockchain to solve a real-world problem, such as supply chain
transparency, the following steps can be taken:
i) Identify the problem: Determine the specific issue within the supply chain
that blockchain can address, such as tracking the origin of products, ensuring
product authenticity, or improving traceability.
ii) Design the system: Develop a blockchain system that includes smart
contracts to automate and enforce transparency in the supply chain. Use a
permissioned blockchain to control access and permissions based on the
different stakeholders involved.
iii) Implement tracking mechanisms: Utilize IoT devices, QR codes, or RFID
tags to track the movement of products along the supply chain. This data is
then recorded on the blockchain to create an immutable record of the
product's journey.
iv) Establish trust and transparency: Allow stakeholders to access the
blockchain to view real-time updates on product movements, verify
authenticity, and ensure compliance with regulations. This transparency
enhances trust among participants and improves the overall integrity of the
supply chain.
v) Continuous monitoring and improvement: Regularly monitor the blockchain
system, analyze data, and make adjustments as needed to optimize
 efficiency, accuracy, and transparency within the supply chain. By
continuously improving the system, the real-world problem can be effectively
addressed using blockchain technology.
CopySearch Web
Delete

Delete
a) Differences between Sequential Consistency and Entry Consistency:
 Sequential Consistency: In sequential consistency, the order of operations on
a single object must be the same for all processes in the system. This means
that each process sees the operations in the same order with respect to that
object.
 Entry Consistency: Entry consistency is a weaker form of consistency where
the order of operations only needs to be consistent within an object (entry).
This means that different processes can have different views of the order of
operations for different objects, as long as the order is consistent within each
object.
b) Differences between Sequential, Causal, and Entry Consistency:
 Sequential Consistency: Requires all processes to agree on the order of
operations on all objects. It ensures that all processes see the same order of
operations, which can lead to delays and inefficiencies in systems with high
contention.
 Causal Consistency: Focuses on preserving causality relationships between
operations. It allows for some operations to be seen out of order as long as
there is no causal dependency between them. This can improve performance
in systems where strict ordering is not necessary.
 Entry Consistency: Provides consistency within individual objects (entries) but
allows for different views of the order of operations across objects. This can
be more efficient for systems where operations on different objects do not
need to be strictly ordered.
From a performance point of view, Entry Consistency would likely be better for
typical business applications operating on sets of data. This is because it
offers more flexibility and can reduce contention and delays by allowing
different views of the order of operations for different objects.
c) Protocols for Achieving Sequential Consistency:
To achieve sequential consistency in a distributed system, protocols such as
Total Order Broadcast (TOB) or Atomic Broadcast can be used. Total Order
Broadcast ensures that all messages are delivered to all processes in the
same order, which helps maintain the order of operations and achieve
sequential consistency across the system. By enforcing a total order of
message delivery, the system can guarantee that all processes see the
operations in the same order, leading to sequential consistency.
d) Consistency in Replicated Databases with Totally Ordered Multicast:
If all operation requests are delivered by totally ordered multicast, the
replicated databases would achieve strong consistency. Totally ordered
multicast ensures that all updates are delivered in the same order to all
replicas, preventing inconsistencies and ensuring that all replicas are up-to-
date and consistent at all times. This level of consistency ensures that all
operations are performed in the same order across all replicas, maintaining
data integrity and avoiding conflicts.
CopySearch Web
Delete

Delete
Facebook is a massive social networking platform that operates as a
distributed system. A distributed system is a collection of independent nodes
that work together to provide a unified service. In the case of Facebook, the
distributed system architecture is crucial for handling the enormous amount of
data, users, and interactions on the platform.
Here are some key aspects of Facebook as a distributed system:
1. Scalability: One of the primary reasons for using a distributed system
architecture is scalability. Facebook serves billions of users worldwide, and a
centralized system would struggle to handle the volume of data and requests.
By distributing its workload across multiple servers and data centers,
Facebook can efficiently scale its infrastructure to meet the demand.
2. Fault Tolerance: Distributed systems are designed to be resilient to failures.
If a single server or data center goes down, the distributed system can
continue to function without significant disruption. Facebook utilizes
redundancy and replication techniques to ensure that user data is safe and
accessible even in the event of hardware failures.
3. Data Replication: To provide users with a seamless experience, Facebook
employs data replication across multiple servers. This ensures that user data
remains consistent and available across different regions. By replicating data,
Facebook can improve performance and reduce latency for users accessing
the platform from different parts of the world.
4. Load Balancing: Facebook uses load balancing techniques to distribute
incoming traffic evenly across its servers. This ensures that no single server is
overwhelmed with requests, optimizing performance and preventing
bottlenecks. Load balancers help Facebook handle spikes in traffic and
maintain a responsive user experience.
5. Caching: To improve performance and reduce the load on backend servers,
Facebook utilizes caching. Commonly accessed data and content are cached
at edge servers or CDN (Content Delivery Network) nodes to minimize latency
and deliver content quickly to users. Caching plays a vital role in enhancing
the user experience on the platform.
6. Consistency and Availability: Maintaining consistency and availability in a
distributed system like Facebook is a complex challenge. Facebook employs
consistency models that balance data integrity with performance and
availability requirements. Techniques such as eventual consistency and
quorum-based systems are used to ensure that data remains consistent
across distributed nodes.
7. Networking Infrastructure: Facebook invests heavily in its networking
infrastructure to support its distributed system. High-speed networking, data
centers located strategically around the globe, and sophisticated routing
protocols enable Facebook to deliver content quickly and reliably to users
worldwide.
In conclusion, Facebook's distributed system architecture plays a crucial role
in delivering a seamless user experience to its global user base. By
leveraging scalability, fault tolerance, data replication, load balancing,
caching, and robust networking infrastructure, Facebook can efficiently handle
the massive scale of data and interactions on its platform.
 CopySearch Web
Delete

Delete
a) Comparison of RPC and RMI with respect to implementation:
1. RPC (Remote Procedure Call):
o Implementation: In RPC, a client program can call procedures on a remote
server as if they were local procedures. This is achieved through stubs and
marshalling. The client-side stub converts the procedure call into network
communication, sends it to the server, which then executes the procedure,
sends the result back, and the stub on the client side unmarshals the result.
2. RMI (Remote Method Invocation):
o Implementation: RMI is specific to Java and enables objects to invoke
methods on remote objects. It uses Java interfaces and classes to define
remote objects and client applications. RMI utilizes Java's serialization to pass
objects by value between the client and server, and stubs play a crucial role in
RMI implementation.
Key Differences:
 RPC is more language-neutral while RMI is Java-specific.
 RMI provides support for passing objects by reference but RPC generally
deals with passing parameters by value.
 In terms of implementation, both RPC and RMI rely on stubs, but RMI typically
has a higher level of abstraction due to its object-oriented nature.
b) Implementation of Logical Clocks in a Distributed System:
In a distributed system, logical clocks help order events that occur in different
processes. The most common implementation of logical clocks is based on
Lamport timestamps, which assign a unique timestamp to each event in the
system. Here's how logical clocks are implemented:
1. Each process in the distributed system maintains its logical clock, initially set
to zero.
2. When an event occurs at a process, the process increments its logical clock
by 1 and assigns the incremented value to the event.
3. When a process sends a message, it includes its current logical clock value in
the message.
4. Upon receiving a message, the receiving process updates its logical clock to
be the max of its current logical clock value and the received timestamp + 1.
5. This way, logical clocks help in establishing the ordering of events in a
distributed system.
c) i) Using Master-Workers Parallelization Pattern for BETTR's Photo
Processing Workload:
 The Master-Workers parallelization pattern involves a master node distributing
tasks to worker nodes, which process the tasks concurrently. In the case of
processing millions of photos per hour, utilizing this pattern can significantly
improve the efficiency and speed of processing by leveraging parallel
computing capabilities.
 I would recommend using the Master-Workers pattern for BETTR's photo
processing workload as it allows for horizontal scaling and efficient distribution
of tasks. By dividing the workload among multiple worker nodes, BETTR can
achieve faster processing times and handle the large demand effectively.
ii) Replication Strategy for Improved Availability in BETTR's Photo Processing
Pipeline:
 Replication involves creating copies of data or services across multiple nodes
to enhance availability and fault tolerance. In the case of BETTR's photo
processing pipeline, replicating critical components such as the image
processing algorithms, databases, and workload distribution mechanisms can
improve reliability and ensure continuous operation in the event of node
failures.
 However, there are potential drawbacks to using replication in this scenario.
Maintaining consistency among replicated data or services can be
challenging, especially in a high-demand environment like photo processing.
Synchronization overhead, network latency, and complexities in handling
concurrent updates across replicas are factors that need to be carefully
considered when implementing replication in BETTR's system.
Overall, while replication can enhance availability, it requires careful planning
and management to address potential drawbacks and ensure the overall
stability and performance of BETTR's photo processing pipeline.
 CopySearch Web
Delete

Delete
a) Issues relating to 'deadlock' in the proposed system:
In the scenario described, where all components of a holiday package must
be successfully booked or none at all, there is a potential for deadlock to
occur. Deadlock is a situation in which two or more processes are unable to
proceed because each is waiting for the other to release a resource, ultimately
leading to a standstill. In the context of allornothing.com, the following
deadlock scenarios may arise:
1. Component Dependencies: If different components of the holiday package
have dependencies on each other, such as a certain flight being required to
book a specific hotel room, there is a risk of deadlock if one component
cannot be booked. For example, if the flight booking fails but the hotel booking
is already reserved, preventing any further progress.
2. Limited Availability: Deadlock may occur if all the desired components have
limited availability and are in high demand. If one component gets booked but
the others are not available, the system may be unable to proceed with the
booking of the remaining components, leading to a deadlock situation.
3. Reservation Durations: If components are kept on hold for a specific
duration while waiting for final payment, deadlock can occur if the hold periods
overlap and users are unable to finalize their bookings within the specified
time frame, causing all components to be released without confirmation.
b) How to avoid deadlock in the proposed system:
To prevent deadlock in the system proposed by allornothing.com, the
following strategies can be implemented:
1. Timeout Mechanism: Implement a timeout mechanism that releases the hold
on components if the final payment is not made within a specified time frame.
This will prevent indefinite waiting and ensure that components are released
for other users if the booking process stalls.
2. Booking Priorities: Establish a prioritization system for booking components
based on dependencies and availability. Ensure that critical components are
booked first to reduce the risk of deadlock due to dependencies on
unavailable resources.
3. Conditional Bookings: Allow users to specify alternative options for each
component in case their first choice is not available. This flexibility can help
avoid deadlock by enabling the system to proceed with alternative bookings if
the primary choices are unavailable.
4. Parallel Processing: Enable concurrent processing of component bookings
to reduce the waiting time and increase the chances of successfully booking
all components. Parallel processing can help avoid situations where one
component blocks the progress of others.
5. Dynamic Resource Allocation: Continuously monitor availability and
dynamically adjust resource allocations to maximize the chances of
successful bookings. This adaptive approach can help mitigate deadlock risks
by proactively managing resource contention.
By incorporating these strategies into the design and implementation of the
allornothing.com system, the likelihood of deadlock occurring can be
minimized, ensuring a smoother and more reliable booking process for users.
CopySearch Web
Delete

Delete
a) i) Non-functional reason for estate agents to participate in the distributed
system:
 I disagree with Jack as the main reason for estate agents to participate in the
distributed system created by Goodlet is functional rather than non-functional.
Estate agents participate in the system primarily to generate leads and
potential customers for their properties, which directly impacts their business
and revenue. By reaching a wider audience through Goodlet's platform, estate
agents have the opportunity to increase the visibility of their properties and
attract more potential buyers, ultimately leading to potential sales and profits.
 ii) Functional reason for estate agents to participate in the distributed system
with a RESTful API:
 By developing a RESTful API to provide access to estate agent data for third-
party applications, Goodlet opens up a new revenue stream by charging these
third-party applications for accessing the valuable property information. This
initiative offers a tangible benefit to estate agents as they stand to receive a
share of the revenue generated through this additional channel. Participating
in the distributed system not only helps estate agents reach potential
customers but also enables them to generate additional income through
revenue sharing, which is a functional reason for their continued participation.
iii) Features of a distributed system in the extended Goodlet system:
 In the extended Goodlet system with a RESTful API for third-party
applications, all components exhibit the essential features of a distributed
system:
1. Multiple Components: The system consists of multiple components such as
the Goodlet platform, estate agent databases, third-party applications, and the
RESTful API.
2. Interconnectivity: These components communicate and interact with each
other over a network, exchanging data and requests through the RESTful API.
3. Concurrency: The system handles concurrent requests from multiple users
and third-party applications accessing estate agent data simultaneously.
4. Fault Tolerance: The system is designed to gracefully handle failures or
disruptions, ensuring that the availability and reliability of services are
maintained.
b) Synchronization in Distributed Systems:
Synchronization in distributed systems refers to the coordination and control
of concurrent processes or components to ensure consistency and avoid
conflicts. In the context of distributed systems, where multiple entities may
operate independently and asynchronously, synchronization mechanisms are
crucial for maintaining data integrity and consistency. Some key aspects of
synchronization in distributed systems include:
1. Consistency Management: Synchronization mechanisms are used to
manage data consistency across distributed nodes. Techniques like
distributed locking, versioning, and distributed transactions help maintain a
consistent view of data across the system.
2. Concurrency Control: Distributed systems often have multiple processes
accessing shared resources concurrently. Synchronization techniques such
as mutual exclusion (e.g., using locks or semaphores) are employed to ensure
 that only one process accesses a critical resource at a time, preventing
conflicts and data corruption.
3. Coherence in Caching: In distributed caching systems, synchronization
mechanisms ensure cache coherence by propagating updates and
invalidations to all nodes that have copies of the cached data. This avoids
inconsistencies that may arise due to stale or outdated data in caches.
4. Event Ordering: Synchronization is essential for establishing a consistent
order of events in distributed systems. Logical clocks, timestamps, and
consensus algorithms help ensure that events are ordered correctly, even in
the presence of network delays and communication issues.
Overall, synchronization plays a critical role in distributed systems by
managing concurrent access, maintaining consistency, and facilitating smooth
collaboration among distributed components.
CopySearch Web
Delete

Delete
a) The concerns of the technical manager regarding the move towards using
web-based distributed computing are indeed well-founded, especially
considering the fact that the existing systems were originally designed with a
focus on CPU efficiency. In a distributed computing environment, emphasis is
often placed on factors such as network latency, data transfer bandwidth, and
parallel processing capabilities rather than just CPU efficiency. As such, the
transition to distributed computing may require significant re-architecting and
optimization of existing systems to ensure they can effectively operate in a
distributed environment. Adapting to distributed computing may involve
redesigning algorithms, data structures, and communication protocols to
account for the complexities introduced by distributed systems, which may
differ from the CPU-centric design principles of the existing systems.
b) Peer-to-peer (P2P) architectures are said to lead naturally to balanced
loads and graceful scaling due to the decentralized nature of the network. In a
P2P system, each node (peer) in the network has equal capabilities and can
 act as both a client and a server, contributing resources and participating in
the network's operations. This leads to several benefits:
 Load Balancing: In a P2P network, tasks and data can be distributed across
multiple nodes, automatically balancing the load and avoiding bottlenecks on
specific nodes. Each peer can contribute resources and share the workload,
leading to a more balanced and efficient system.
 Scalability: P2P networks can scale gracefully as new nodes can easily join
or leave the network without affecting the overall system's performance. The
distributed nature of P2P systems allows them to handle increasing loads and
accommodate a growing number of participants without centralized
constraints.
c) The Two Generals problem is an unsolvable issue in distributed computing
where two communication-restricted generals must agree on a coordinated
attack plan. The assertion that it is impossible to design an algorithm
guaranteed to lead to a coordinated solution stems from the inherent
limitations of asynchronous communication and the impossibility of achieving
absolute certainty in distributed systems. In scenarios like the Two Generals
problem, where messages can be lost, delayed, or delivered out of order,
there is no foolproof method to ensure that both parties have reached a
definitive agreement without any ambiguity or risk of failure. As a result,
designing an algorithm that guarantees a coordinated solution in the face of
these challenges is fundamentally impossible.
d) Benefits of adopting grid computing for a university:
 Resource Optimization: Grid computing allows universities to efficiently
utilize and share resources such as computing power, storage, and research
data across departments and research projects. This optimizes resource
usage and maximizes productivity.
 Collaboration and Research: Grid computing facilitates collaboration among
researchers, enabling them to access shared resources and work together on
complex scientific and academic projects. This enhances research capabilities
and accelerates progress in various fields.
 Scalability: Grid computing provides scalability to accommodate growing
computational needs and handle large-scale simulations, data analysis, and
processing requirements. Universities can easily scale up their computing
infrastructure to meet evolving demands.
 Cost Efficiency: By sharing resources and infrastructure through grid
computing, universities can reduce costs associated with maintaining and
managing individual computing systems. This cost efficiency allows
 institutions to allocate resources effectively to support research and academic
activities.
 High Performance Computing: Grid computing offers universities access to
high-performance computing capabilities, allowing them to tackle
computationally intensive tasks and simulations that require substantial
computational power. This enables researchers to carry out advanced
analyses and simulations efficiently.
CopySearch Web
Delete

Delete
a) Measures that Google has in place to ensure transparency and the type of
Distributed System transparency they are enforcing:
1. Service Level Agreements (SLAs): Google provides clear SLAs to its
customers, outlining the level of service and availability they can expect. This
ensures transparency regarding the performance metrics and service
guarantees. This relates to Behavioral Transparency in Distributed Systems,
where the behavior of the system is clear and predictable.
2. Monitoring and Logging: Google utilizes monitoring and logging tools to
track the performance of their services and provide insights into system
operations. This transparency mechanism falls under Communication
Transparency, ensuring that system components can communicate
effectively and securely.
3. Data Protection and Privacy Policies: Google follows strict data protection
and privacy policies, detailing how user data is collected, stored, and used. By
making these policies transparent to users, Google enforces Access
Transparency, ensuring that users have visibility into the handling of their
data.
4. Security Audits and Compliance: Google undergoes regular security audits
and complies with industry regulations to maintain a secure environment for
their services. This reflects Location Transparency in Distributed Systems,
 where users do not need to be aware of the physical location of resources for
access.
b) Examples of services offered by Google based on the models:
 Software as a Service (SaaS): Google offers services like GMail, Google
Docs, and Heroku as examples of Software as a Service where users can
access software applications over the internet without the need for local
installation.
 Platform as a Service (PaaS): Google App Engine (GAE) is an example of
Platform as a Service, providing a platform for developers to build, deploy,
and scale applications without managing the underlying infrastructure.
 Infrastructure as a Service (IaaS): Google Compute Engine (GCE) and
Google Kubernetes Engine (GKE) are examples of Infrastructure as a
Service, allowing users to create virtual machines and manage containers in a
flexible and scalable infrastructure.
c) Possible models to describe Google's system model:
1. Cloud Computing Model: Google's services align with a cloud computing
model, where resources are provided as services over the internet. This
model allows for on-demand access to computing resources, scalability, and
flexibility.
2. Microservices Architecture Model: Google's use of containers, Kubernetes,
and modular services reflects a microservices architecture model. This
approach enables flexibility, scalability, and independent deployment of
services.
3. Platform-based Ecosystem Model: Google's range of services, including
software, platform, and infrastructure offerings, form an ecosystem that caters
to developers, businesses, and end-users. This model emphasizes
interconnectivity, integration, and ease of use across the Google ecosystem.
d) Premises of distributed systems explained in the context of Google:
1. Concurrency: Google's system supports multiple users and services
simultaneously, handling numerous requests and operations concurrently to
ensure efficient use of computing resources.
2. Scalability: Google's services can scale to accommodate increasing demand,
allowing for the seamless expansion of resources and capabilities as needed.
3. Fault Tolerance: Google implements redundancy, backup systems, and
failover mechanisms to ensure service availability and uptime, even in the
face of hardware failures or disruptions.
4. Openness: Google's services are accessible to a wide range of users and
developers, with open APIs, documentation, and collaboration opportunities
promoting openness and interoperability.
e) How Google manages key aspects in the implementation of their services:
i. Openness: Google promotes openness through open APIs, developer
tools, and documentation, allowing users and developers to access and
integrate with their services transparently. ii. Heterogeneity: Google handles
heterogeneity by supporting diverse platforms, devices, and technologies
within their ecosystem, ensuring compatibility and interoperability across
different systems. iii. Fault Tolerance: Google employs redundancy, load
balancing, and fault-tolerant architectures to mitigate the impact of failures
and ensure continuous service availability and reliability. iv. Concurrency:
Google manages concurrency by efficiently handling multiple requests and
operations simultaneously, utilizing parallel processing and distributed
computing techniques to maximize performance.
f) How Google implements types of distributed systems: i. Distributed
Information System: Google utilizes distributed databases, indexing, and
caching mechanisms to provide fast and reliable access to information across
its services, ensuring data availability and consistency. ii. Pervasive
Systems: Google's services extend beyond traditional computing devices,
supporting internet-connected home appliances, mobile devices, and
wearable technology, creating a pervasive ecosystem that integrates
seamlessly into users' everyday lives.
CopySearch Web
Delete

Delete
a) Transparency in distribution refers to the ability of a distributed system to
hide the complexities of its underlying architecture and provide a seamless
user experience. There are various types of transparency in distributed
systems:
 Access Transparency: Users can access resources in a distributed system
without needing to know the physical location or distribution of those
resources.
 Location Transparency: Users do not need to be aware of the physical
location of resources in a distributed system, allowing for seamless access
regardless of where resources are located.
 Migration Transparency: The movement of resources between different
locations in a distributed system is transparent to users and applications.
 Relocation Transparency: Resources can be relocated within a distributed
system without affecting user access or application functionality.
 Replication Transparency: The replication of data or resources in a
distributed system is transparent to users, helping to enhance availability and
reliability.
 Failure Transparency: Failures or disruptions in the system are handled
transparently, ensuring that users experience minimal impact.
 Concurrency Transparency: Users can perform concurrent operations in a
distributed system without needing to manage the complexities of
synchronization and communication between multiple processes.
b) Definitions and examples of how Hadoop enforces features of distributed
systems: i. Distributed file storage: Distributed file storage refers to the
storage of files across multiple nodes in a distributed system, enabling
reliability, scalability, and fault tolerance. Hadoop enforces distributed file
storage through the Hadoop Distributed File System (HDFS), which stores
large files across multiple data nodes in a cluster. HDFS replicates data
blocks across nodes to ensure fault tolerance and availability. For example,
when a file is uploaded to HDFS, it is split into blocks and distributed across
nodes in the cluster for parallel processing.
ii. Data locality: Data locality is the principle of processing data where it
resides to minimize data movement and improve performance in distributed
systems. Hadoop enforces data locality by scheduling tasks to run on nodes
where data is stored, reducing network traffic and improving processing
efficiency. For example, when running a MapReduce job in Hadoop, tasks are
scheduled to process data blocks that are stored locally on the same node,
maximizing data locality and minimizing data transfer across the network.
iii. Parallel processing of data: Parallel processing of data involves splitting
data into smaller chunks and processing them simultaneously across multiple
nodes in a distributed system to achieve faster processing times. Hadoop
enforces parallel processing through its MapReduce framework, which divides
 data processing tasks into map and reduce phases that can run in parallel
across nodes in a cluster. For example, when executing a MapReduce job in
Hadoop, data is processed in parallel across multiple nodes, allowing for
efficient computation and analysis.
iv. Fault tolerance: Fault tolerance in distributed systems refers to the
system's ability to continue functioning and serving users in the event of
hardware failures, network interruptions, or other issues. Hadoop enforces
fault tolerance through data replication and job recovery mechanisms. For
example, in HDFS, data blocks are replicated across multiple nodes to ensure
data availability in case of node failures. Additionally, the MapReduce
framework in Hadoop can restart failed tasks on other nodes to complete
processing jobs even in the presence of failures.
CopySearch Web
Delete

Delete
Remote Procedure Call (RPC) is a communication protocol that allows a
program to request a service from a program located on a different computer
in a network. RPC can be used to ensure different types of transparency in a
distributed system, including migration, replication, and failure transparency:
1. Migration Transparency: Migration transparency refers to the ability of a
distributed system to move resources or services from one location to another
without impacting users or applications. RPC can be used to achieve
migration transparency by abstracting the underlying details of the migration
process from the users or applications. When a service is migrated to a
different server or location, RPC allows the client program to continue making
requests to the service without needing to know the new location. The RPC
framework handles the communication between the client and the migrated
service transparently. This ensures that users or applications can access the
service seamlessly, regardless of its physical location.
2. Replication Transparency: Replication transparency involves duplicating
data or services across multiple nodes in a distributed system to improve
 availability and reliability. RPC can be used to ensure replication transparency
by abstracting the replication process from users or applications. When a
service is replicated on multiple servers, RPC allows clients to access the
service without needing to know which instance of the service they are
communicating with. The RPC framework manages the communication
between the client and the replicated services transparently, ensuring that
requests are distributed evenly across the replicas. This enables users or
applications to access the service seamlessly, even in the presence of
replication.
3. Failure Transparency: Failure transparency refers to the ability of a
distributed system to handle failures or disruptions without impacting users or
applications. RPC can be utilized to ensure failure transparency by providing
mechanisms for fault tolerance and recovery. In the event of a failure, such as
a server crash or network issue, RPC can handle the failure gracefully by
rerouting requests to other available servers or nodes. The RPC framework
can automatically detect failures and redirect requests to alternative
resources, ensuring that users or applications experience minimal disruption.
By abstracting the details of fault tolerance from users or applications, RPC
enables the system to maintain availability and reliability in the face of failures.
CopySearch Web
Delete

Delete
a)
1. Access Transparency: Users can access resources in a distributed system
without needing to know the physical location or distribution of those
resources. This transparency allows users to interact with resources as if they
were all located locally, regardless of their actual location.
2. Location Transparency: Users do not need to be aware of the physical
location of resources in a distributed system, allowing for seamless access
regardless of where resources are located. This transparency enables the
 system to manage resource placement and movement without affecting user
interactions.
3. Migration Transparency: The movement of resources between different
locations in a distributed system is transparent to users and applications. This
transparency allows resources to be relocated or replicated without disrupting
user access or application functionality.
4. Replication Transparency: The replication of data or resources in a
distributed system is transparent to users, helping to enhance availability and
reliability. This transparency allows the system to maintain multiple copies of
data or services to improve fault tolerance and performance without users
needing to manage the replication process.
b) i. Distributed systems versus parallel systems:
 Distributed systems involve multiple interconnected computers that
communicate and coordinate with each other to achieve a common goal.
These systems typically span across different locations and are designed to
work together to solve complex problems. Parallel systems, on the other
hand, involve multiple processing units within a single computer or closely
connected computers that work together to execute tasks simultaneously. The
main difference is that distributed systems focus on sharing resources and
collaborating over a network, while parallel systems focus on utilizing multiple
processors for simultaneous computation on a single machine or tightly
coupled machines.
ii. Remote method invocation versus Remote procedure call:
 Remote method invocation (RMI) is a Java-based technology that allows a
Java object to invoke methods on an object running in another JVM, usually
on a different machine. RMI is specific to Java and enables objects to
communicate and interact across a network. Remote procedure call (RPC), on
the other hand, is a generic protocol that allows a program to execute
procedures or functions on a remote computer. RPC is language-independent
and can be used to implement communication between different programming
languages or systems.
iii. Synchronous versus asynchronous communication:
 Synchronous communication is a communication method where the sender
waits for a response from the receiver before proceeding with other tasks. In
this mode of communication, both the sender and receiver are engaged in a
real-time conversation. Asynchronous communication, on the other hand,
allows the sender to send a message to the receiver without waiting for an
immediate response. The sender can continue with other tasks, and the
 receiver processes the message at its own pace. Asynchronous
communication is typically used in scenarios where immediate responses are
not required or when there may be delays in message processing.
CopySearch Web
Delete

Delete

Delete
b) Differences between persistent and transient messaging:
1. Durability:
 Persistent messaging ensures that messages are stored in a stable storage
system and are not lost even in the event of system failures or crashes. This
means that messages can be recovered and delivered at a later point in time,
providing reliability and consistency in message delivery. On the other hand,
transient messaging does not guarantee message durability, as messages are
typically held in memory and may be lost in case of system failures.
2. Message Lifetime:
 Messages in persistent messaging systems persist until they are explicitly
acknowledged or processed by the receiver, or until a predefined expiration
time. This allows messages to be stored and retrieved even after extended
periods. In transient messaging, messages are often considered transient and
are discarded after they have been processed or if they expire, making them
short-lived in nature.
c) Role of Middleware in a Distributed System: Middleware acts as a
software layer that provides communication and coordination services to
facilitate interaction between different components in a distributed system. It
abstracts the complexities of communication protocols, hardware differences,
and data formats, allowing different applications and systems to seamlessly
communicate with each other. The diagram below illustrates the role of
middleware in a distributed system:

Question4

CBA Systems is a software development company that develops software specifically for universities.
You are required to describe in terms of functionality and use a suggested application that CBA
holdings can develop to serve the needs of universities which is modeled according to the following
distributed system type:

a) Cluster computing system

b) Grid computing system

[20)