MODULE – III

Cloud Platform Architecture over Virtualized Datacenters

Syllabus: Cloud Platform Architecture over Virtualized Datacenters: Cloud

Computing and Service Models, Data Center Design and Interconnection
Networks, Architectural Design of Compute and Storage Clouds, Public Cloud
Platforms: GAE, AWS and Azure, Inter-Cloud Resource Management
[Textbook 1 Chapter 4: 4.1 to 4.5]

4.1 CLOUD COMPUTING AND SERVICE MODELS

 Over the past two decades, the world economy has rapidly moved from
manufacturing to more service-oriented. In 2010, 80 percent of the U.S. economy was
driven by the service industry, leav ing only 15 percent in manufacturing and 5 percent
in agriculture and other areas. Cloud computing benefits the service industry most and
advances business computing with a new paradigm.
 In 2009, the global cloud service marketplace reached $17.4 billion. IDC predicted in
2010 that the cloud based economy may increase to $44.2 billion by 2013. Developers
of innovative cloud applications no longer acquire large capital equipment in advance.
They just rent the resources from some large data centers that have been automated
for this purpose.
 Users can access and deploy cloud applications from anywhere in the world at very
competitive costs. Virtualized cloud platforms are often built on top of large data
centers. With that in mind, we examine first the server cluster in a data center and its
interconnec tion issues.
 In other words, clouds aim to power the next generation of data centers by
architecting them as virtual resources over automated hardware, databases, user
interfaces, and application envir onments. In this sense, clouds grow out of the desire
to build better data centers through automated resource provisioning.
 Public, Private, and Hybrid Clouds

1
 Cloud computing has evolved from cluster, grid, and utility computing. Cluster and
grid computing use multiple computers in parallel to solve problems of various sizes.
Utility computing and Software as a Service (SaaS) introduce the model of pay-per-
use computing resources.
 Cloud computing is a high-throughput computing (HTC) paradigm that delivers
services through large data centers or server farms. It allows users to access shared
resources from anywhere at any time using connected devices. Cloud computing
moves computations to the location of data rather than transferring large datasets to
multiple desktops, thus achieving better network bandwidth utilization.
 With machine virtualization, cloud computing achieves better resource utilization,
application flexibility, and cost efficiency. It uses a virtual platform with elastic
resources, provisioning hardware, software, and datasets dynamically on demand.
The cloud model replaces desktop computing with a service-oriented platform
supported by server clusters and large databases.
 Cloud computing helps IT companies by eliminating the need to manage hardware
and system software. It enables developers to focus on application development and
business value rather than infrastructure setup. It offers simplicity and low cost to
both providers and end users.
 Centralized versus Distributed Computing

• Some believe cloud computing is centralized in data centers.

• Others argue it is distributed computing using data center resources.
• How Cloud Computing Works
• Cloud applications distribute tasks to servers in data centers.
• These servers are virtual machines (VMs) in virtual clusters.
• Cloud platforms are distributed through virtualization.
• Geographical Distribution
• Cloud providers distribute their data centers globally.
• Helps with fault tolerance, faster response times, and legal compliance.
• Challenges in Cross-Border Cloud Usage
• Users in different countries may hesitate to use foreign cloud services.
• Service-Level Agreements (SLAs) are needed for trust and security.

 Public Clouds

• Built over the Internet and accessible to paid users.

• Owned by service providers and available through subscriptions.
• Examples of Public Clouds

o Google App Engine (GAE)

o Amazon Web Services (AWS)
o Microsoft Azure
o IBM Blue Cloud
o Salesforce.com’s Force.com

2
 Private Clouds

o Built within an intranet and owned by a single organization.

o Not publicly accessible; used by the organization and its partners.
o Client-owned and managed for internal use.
o Purpose and Benefits
o Provides a flexible and agile infrastructure for internal workloads.
o Ensures greater customization and organizational control.
o Delivers efficient and convenient cloud services.
o Ex: IBM RC2 –Developed by IBM to handle multiple research centers

 How Public Clouds Work

• Providers offer a remote interface for users to create and manage VM instances.
• Operates on proprietary infrastructure managed by the provider.
• Delivers business processes, applications, and infrastructure as services.
• Uses a flexible, pay-per-use pricing model.
• Ex: Netflix uses AWS for streaming videos

 Hybrid Clouds

• A combination of both public and private clouds.

• Allows organizations to use private infrastructure while accessing public cloud
resources when needed.

 How It Works
1. Private clouds supplement their capacity by integrating with a public cloud.
2. Example: Flipkart during BB SALES: Uses private cloud: storing customer data
& Public cloud:AWS for scaling resources during peak demands.
3. Example: Research Compute Cloud (RC2) is a private cloud, built by IBM, that
interconnects the computing and IT resources at eight IBM Research Centers
scattered throughout the United States, Europe, and Asia

 Who Can Access Hybrid Clouds?

1. Clients (internal users of the organization).
2. Partner networks (business partners and suppliers).
3. Third parties (external users with restricted access).
 Advantages of Hybrid Clouds

1. Balances standardization & customization – Uses public clouds for flexibility while
keeping sensitive data private.
2. Preserves capital investment – Organizations don’t have to expand private
infrastructure but can use public cloud resources when needed.
3. Enhances security & efficiency – Keeps critical operations private while leveraging
public cloud scalability.

3
  Data-Center Networking Structure

 Core Components of a Cloud Data Center

• Server Cluster (or VM Cluster) – The main computing units of the cloud.
• Compute Nodes – Servers performing user jobs.
• Control Nodes – Manage and monitor cloud activities.
• Gateway Nodes – Act as access points for external users and ensure security.

 Cloud Networking

• Layer 2 (Bottom Layer) –

 Contains racks of servers (A).
 Servers are connected via Layer 2 switches (S).
• Layer 3 (Top Layer) –
 Connects data centers to the Internet.
 Uses Access Routers (AR) and Border Routers (BR) for network traffic management.
• Load Balancers (LB) –
 Distribute workloads efficiently across multiple servers.
 Unlike traditional grids with static resource demands, cloud systems are designed to
handle fluctuating workloads by providing variable resources dynamically. Private
clouds can effectively handle this if they are well-designed and managed.

 Examples of Private Clouds

• NASA's Cloud – Supports climate modeling, reducing costs for researchers.

• CERN’s Cloud – Distributes data and computing resources to scientists worldwide.

4
  Different clouds require varied performance, security, and data protection levels.
 SLAs help ensure both providers and users agree on service expectations.

 Comparison: Data Centers vs. Supercomputers

• Data centers and supercomputers share some similarities but have fundamental
differences. Data centers scale horizontally, with server clusters often built from
thousands to millions of nodes.
• For instance, Microsoft's data center in Chicago houses 100,000 eight-core servers.
Supercomputers use separate data farms, while data centers rely on disks on server
nodes, memory cache, and databases.
• Supercomputers use high-bandwidth custom networks like fat trees or 3D torus
networks. In contrast, data centers use IP-based commodity networks, such as 10 Gbps
Ethernet, optimized for Internet access

 Cloud Development Trends

 Private clouds & Hybrid Clouds

Private clouds are expected to grow faster than public clouds.

• They offer better security and trust within organizations.

• This means the boundary between public and private clouds may blur in the future.
• Hybrid clouds will likely dominate the cloud landscape..

 Types of Service Nodes in Cloud Computing

• Runtime Supporting Service Nodes – Help cloud clusters function smoothly

• Independent Service Nodes – Provide standalone services

Ex:Uber App

RSN:ride requests and ensure smooth communication between drivers and passengers.

ISN:provide real-time traffic updates and weather conditions for better route selection.

 Performance and Efficiency in Cloud Computing

• Clouds reduce large data movement, lowering Internet traffic.

• Cost-effective performance is a key focus in cloud computing.

 Future Cloud Performance Considerations

• Future research will focus on:

• Performance modelling
• Service availability & fault tolerance
• Data protection & security
• Operating cost optimization

5
 Cloud Ecosystem and Enabling Technologies

 Classical Computing (Traditional Model)

• Buy and Own – Organizations must purchase hardware, system software, and
applications.
• High Maintenance –
• Install, configure, test, verify, evaluate, and manage the system.
• Requires IT staff and infrastructure management.
• Fixed Costs – Pay for maximum capacity, even if it's not used all the time.
• Obsolescence Cycle – Every 18 months, hardware becomes outdated, requiring
costly upgrades.
• High Cost ($$$$$) – Large upfront capital investment is required.

 Cloud Computing (Modern Model)

• Pay-As-You-Go – Subscribe to services instead of owning hardware.

• No Installation or Management – Cloud provider handles infrastructure setup and
maintenance.
• Usage-Based Pricing – Pay only for what is used, reducing costs.
• Cost Savings (80-95%) –
• No upfront capital investment.
• Lower IT staff requirements.
• More flexible and scalable.
• Ideal for Small Businesses –
• No need for expensive servers or frequent upgrades.
• Easy access to computing resources as needed.

 Motivation Behind Cloud Computing

• Cost efficiency – Eliminates large upfront costs and maintenance.

• Scalability – Resources can be scaled up/down as needed.
• Accessibility – Services can be accessed from anywhere via the Internet.
• Flexibility – Businesses can use computing power without owning hardware.

6
  Cloud Design Objectives (Shift Services, Scale Data, High Standards)

Despite the controversy surrounding the replacement of desktop or deskside computing by

centra lized computing and storage services at data centers or big IT companies, the cloud
computing com munity has reached some consensus on what has to be done to make cloud
computing universally acceptable. The following list highlights six design objectives for cloud
computing:

• Shifting computing from desktops to data centers :Computer processing, storage, and
software delivery is shifted away from desktops and local servers and toward data centers
over the Internet

. • Service provisioning and cloud economics: Providers supply cloud services by signing SLAs
with consumers and end users. The services must be efficient in terms of computing, storage,
and power consumption. Pricing is based on a pay-as-you-go policy.

• Scalability in performance :The cloud platforms and software and infrastructure services
must be able to scale in performance as the number of users increases. 198 CHAPTER 4 Cloud
Platform Architecture over Virtualized Data Centers

• Data privacy protection: Can you trust data centers to handle your private data and
records? This concern must be addressed to make clouds successful as trusted services.

• High quality of cloud services: The QoS of cloud computing must be standardized to make
clouds interoperable among multiple providers.

• New standards and interfaces: This refers to solving the data lock-in problem associated
with data centers or cloud providers. Universally accepted APIs and access protocols are
needed to provide high portability and flexibility of virtualized applications.

 Cost Model

Traditional IT Cost Model (Figure 4.3a)

✅ High Fixed Costs – Requires purchasing computers and infrastructure upfront.

✅ Operational Expenses – Includes maintenance, personnel, and service costs.
✅ Costs Increase Rapidly – As users grow, operational expenses rise significantly.
✅ Less Flexibility – Hardware investments limit scalability.

7
Cloud Computing Cost Model (Figure 4.3b)

✅ No Upfront Cost – No need to buy hardware; only pay for usage.

✅ Only Variable Costs – Pay-per-use model, reducing unnecessary expenses.
✅ Highly Scalable – Costs grow gradually with users, making it budget-friendly.
✅ Better for Startups – Removes financial burden by avoiding large capital investments.

 Cloud Ecosystems

The cloud ecosystem refers to the interconnected network of cloud services,

providers, users, and technologies that work together to deliver cloud computing
solutions.

 With the growth of Internet clouds, a broad ecosystem of providers, users, and
technologies has emerged. This ecosystem is primarily built around public clouds, but
there's a growing interest in open-source cloud tools that help organizations
construct their own IaaS (Infrastructure as a Service) clouds using internal
infrastructure. Private and hybrid clouds also leverage public cloud elements,
allowing remote access through web service interfaces such as Amazon EC2.
 Cloud Ecosystem Levels

 Cloud Consumers (Users) – (Level a)

• Need a flexible platform for raw infrastructure, outsourcing, and services.

• Examples: Individual users, Other clouds, Platform-as-a-service (PaaS).

 Cloud Management – (Level b)

• Manages virtualized resources over an IaaS platform.

• Example cloud tools: Amazon EC2, Eucalyptus, Globus Nimbus.
• Uses cloud interfaces (Amazon EC2WS, Nimbus WSRF, ElasticHosts REST).

 VI Management (Virtual Infrastructure Management) – (Level c)

• Allocates virtual machines (VMs) across multiple server clusters.

• Example tools: OpenNebula, VMware vSphere.

8
  VM Management – (Level d)

• Handles VMs installed on servers.

• Example tools: Xen, KVM, VMware.

• Startups & enterprises are shifting to cloud-based IT strategies.

• Private/hybrid clouds offer flexibility & remote access.
• VI management tools support dynamic scaling, load balancing, and server
consolidation.

 Cloud Tools & Interfaces

🔹 Public clouds: Amazon EC2.

🔹VI management: OpenNebula, VMware vSphere.

🔹 Cloud management: Eucalyptus, Globus Nimbus

🔹 VM management: Xen, KVM, VMware.

Many startups are moving to cloud-based IT strategies, reducing capital expenses by avoiding
the setup of dedicated IT infrastructure. This shift pushes the need for a flexible and open
cloud architecture that supports the construction of private/hybrid clouds. VI management
is central to this goal.

Examples of VI management tools include oVirt, vSphere/4 (from VMware), and VM

Orchestrator (from Platform Computing). These tools offer:

 Dynamic VM placement
 Automatic load balancing
 Server consolidation
 Dynamic resizing and partitioning of infrastructure

Besides commercial clouds like Amazon EC2, open-source tools such as Eucalyptus and
Globus Nimbus are available for cloud infrastructure virtualization. Access to these tools is
provided through interfaces like:

 Amazon EC2WS
 Nimbus WSRF
 ElasticHost REST

For VM generation and control, tools like OpenNebula and VMware vSphere support
virtualization platforms such as Xen, KVM, and VMware.

 Infrastructure-as-a-Service (IaaS)

 Cloud Computing Service Models

 Cloud computing offers three main service models:
🔹 IaaS (Infrastructure-as-a-Service)

9
 🔹 PaaS (Platform-as-a-Service)
🔹 SaaS (Software-as-a-Service)

• These models follow a pay-as-you-go approach, making IT resources more

affordable and scalable.
• Service Level Agreements (SLAs)
• Cloud services are provided based on agreements between providers & users.
• SLAs ensure availability, performance, data protection, and security.

 IaaS – Infrastructure as a Service

• Users rent virtualized IT resources (computing, storage, networking).

• No need for hardware investment; resources are on-demand & scalable.
• Users deploy their own OS & applications while the cloud provider manages the
infrastructure.

 IaaS Components

• Compute Instances as a Service (e.g., Amazon EC2, Google Compute Engine).

• Storage as a Service (e.g., Amazon S3, Google Cloud Storage).
• Networking as a Service (e.g., Virtual Private Cloud - VPC).

 IaaS Providers & Examples

• Amazon Web Services (AWS) – EC2 (compute), S3 (storage).

• Microsoft Azure – Virtual Machines, Blob Storage.
• Google Cloud Platform (GCP) – Compute Engine, Cloud Storage.
• Other providers – GoGrid, FlexiScale, Aneka.

 User Control in IaaS

• Users control OS, storage, and applications.

• Providers manage the hardware & physical infrastructure.

Example: Amazon EC2 & S3:A startup uses Amazon EC2 for on-demand computing power
and S3 for cloud storage, reducing costs and improving flexibility.

10
 Platform-as-a-Service (PaaS)

 What is PaaS?

• A cloud-based platform for developing, deploying, and managing applications.

• Includes OS, runtime libraries, and development tools.
• Developers focus on writing code while the provider manages the infrastructure.

 Features of PaaS

• Pre-configured development environments (e.g., Java, Python, .NET).

• Supports collaborative development (global teams can work on the same app).
• Reduces the need for in-house hardware & software infrastructure.

 PaaS Examples

• Google App Engine – Deploys apps in a fully managed environment.

• Microsoft Azure App Services – Hosts web & mobile applications.
• AWS Elastic Beanstalk – Simplifies application deployment.
• Heroku – A cloud platform supporting multiple programming languages.

11
 Example : Google App Engine for PaaS Applications
• Platform as a Service (PaaS): GAE is a cloud-
based platform that allows developers to build
and deploy applications without worrying about
infrastructure.
• Automatic Scaling & Load Balancing: GAE
automatically manages application scaling based
on traffic demands and distributes the load
among multiple servers.
• Task Scheduling: GAE has a distributed
scheduling mechanism to trigger tasks at
specified times and intervals.
• Local Development Environment: Developers
can build, test, and debug applications on their
local machines before deploying them to
Google’s cloud infrastructure.

 Software as a Service (SaaS)

 What is SaaS?

Software as a Service (SaaS) is a cloud-based software delivery model where applications

are hosted online and accessed via a web browser instead of being installed on a personal
computer.

• No need for upfront investment in servers or software.

• Easily accessible from anywhere with an internet connection.
• Managed by the provider, meaning automatic updates and maintenance.

 How SaaS Works?

o A user opens a web browser and logs into a SaaS application (e.g., Gmail).
o The application runs on cloud servers, not on the user's device.
o Data is stored in the cloud, either in a vendor’s proprietary cloud (Google Drive,
Microsoft OneDrive) or a publicly hosted cloud.

 Examples of SaaS Applications

• Email Services: Google Gmail, Microsoft Outlook

• Collaboration Tools: Google Docs, Microsoft SharePoint
• Customer Relationship Management (CRM): Salesforce
• Streaming Services: Netflix, Spotify

12
4.2 DATA-CENTER DESIGN AND INTERCONNECTION NETWORKS

 Warehouse-Scale Data-Center Design

Large-Scale (Warehouse)
Feature Small Modular Data Centers
Data Centers

As large as a shopping
Size Fits inside a 40-ft truck container
mall (11x football field)

Number of 400,000 – 1 million

Around 1,000 servers
Servers servers

Highly scalable (can add

Scalability Limited scalability (fixed capacity)
more servers easily)

Cost Lower per-unit cost due

Higher per-unit cost (more expensive to operate)
Efficiency to economies of scale

Network $13 per Mbps (cheaper

7 times higher than large data centers
Cost per unit)

Storage Cost $0.4 per GB 5.7 times higher than large data centers

Fixed location (requires

Deployment Portable (easily moved and deployed)
infrastructure setup)

Cloud computing, AI, big Temporary setups, remote locations, disaster

Use Case
data, streaming services recovery

Google, Microsoft,
Example Edge computing units, emergency response centers
Amazon data centers

 Data-Center Construction Requirements

 Most data centers are built using commercially available components. An off-the-shelf
server typically includes multicore CPUs with internal cache hierarchies, local shared
DRAM, and directly attached disk drives. These servers are connected via first-level rack
switches, and the entire rack structure is linked using a cluster-level switch. For example,
in a data center with 2,000 servers, each having 8 GB DRAM and four 1 TB disks, groups
of 40 servers are connected through 1 Gbps links to rack-level switches, which are in turn
connected to the cluster-level network.
 There is a significant performance difference between local and off-rack storage. Local
disks provide 200 MB/s bandwidth, while off-rack disks offer only 25 MB/s due to shared
uplinks. Moreover, the total disk storage in such clusters is millions of times greater than

13
 local DRAM capacity, making latency, bandwidth, and capacity management a major
challenge for large applications.
 In large-scale data centers, components are cheaper, but failure is common—about 1%
of nodes may fail concurrently, either due to hardware issues like CPU, disk I/O, or
network failure, or due to software bugs. In severe cases, even the entire data center
may go down, such as during a power outage.
 To ensure reliability, redundant hardware is used, and software must maintain multiple
data copies across different locations. This redundancy ensures that services and data
remain accessible even in the face of hardware or software failures

 Cooling System of a Data-Center Room

Figure 4.9 shows the layout and cooling facility of a warehouse in a data center.

 The data-center room has raised floors for hiding cables, power lines, and cooling
supplies. The cooling system is somewhat simpler than the power system. The raised
floor has a steel grid resting on stanchions about 2–4 ft above the concrete floor. The
under-floor area is often used to route power cables to racks, but its primary use is to
distribute cool air to the server rack.
 The CRAC (computer room air conditioning) unit pressurizes the raised floor plenum
by blowing cold air into the plenum. The cold air escapes from the plenum through
perforated tiles that are placed in front of server racks. Racks are arranged in long
aisles that alternate between cold aisles and hot aisles to avoid mixing hot and cold
air.
 The hot air produced by the servers circulates back to the intakes of the CRAC units
that cool it and then exhaust the cool air into the raised floor plenum again. Typically,
the incoming coolant is at 12–14°C and the warm coolant returns to a chiller.
 Newer data centers often insert a cooling tower to pre-cool the condenser water
loop fluid. Water-based free cooling uses cooling towers to dissipate heat. The cooling
towers use a separate cooling loop in which water absorbs the coolant’s heat in a
heat exchanger

14
  Data-Center Interconnection Networks

A critical core design of a data center is the interconnection network among all servers in the
data center cluster. This network design must meet five special requirements: low latency,
high band width, low cost, message-passing interface (MPI) communication support, and
fault tolerance.

 Application Traffic Support

Data center networks must support diverse traffic patterns, especially MPI (Message
Passing Interface) communication for parallel applications. This includes both:

 Point-to-point communication (between two nodes).

 Collective communication (like broadcasting or gathering across many nodes).

For example:

 Distributed file systems use one-to-many communication, where metadata servers

communicate with many slave servers.
 MapReduce operations require the network to efficiently handle large-scale data
movement during map and reduce phases.

Therefore, the network topology must ensure high bisection bandwidth, so no part of the
cluster becomes a traffic bottleneck.

 Network Expandability

Data centers often grow over time. The interconnection network must be:

 Scalable to support thousands (or even hundreds of thousands) of nodes.

 Flexible to allow restructuring of topology as servers are added.

Key concerns:

 Avoiding bottlenecks as server numbers grow.

 Using low-cost Ethernet switches in topologies like fat-tree or crossbar, though
scalability becomes a challenge at massive scales.
 Supporting modular growth using data-center containers (portable, pre-configured
server modules).

Containers can be added easily: just plug in power, cooling, and network, making
expansion more cost-effective and efficient.

 Fault Tolerance and Graceful Degradation

Networks must tolerate failures and continue running smoothly even when:

 Switches or links fail.

 Servers crash (1% failure rate is common in large data centers).

15
Mechanisms for fault tolerance:

 Multiple network paths between any two servers.

 Redundant servers and data replication.
 Software-level awareness to avoid failed links.
 Hot-swappable hardware components (can be replaced without shutting down the
system).

️ The design must avoid single points of failure and ensure graceful degradation—
continued operation even when parts of the system fail.

 Switch-Centric Data-Center Design

There are two main design philosophies:

 Switch-centric: Standard switches connect servers; no changes needed to servers.

 Server-centric: Servers help route traffic using custom drivers or OS modifications.

Switch-centric design is more traditional and easier to implement at scale because:

 It isolates network logic from server hardware and software.

 Reduces complexity in managing server operations.

The switch organization remains essential to ensure efficient traffic flow regardless of the
approach.

Example 4.4 A Fat-Tree Interconnection Network for Data Centers

 Structure

• Two-layer topology.

16
 • Bottom layer: Server nodes connected to edge switches.
• Upper layer: Aggregation switches connecting edge switches.
• Core switches connect all pods together.
• Pods:
• Core Switches:
• Provide interconnection between pods.
• Enable communication across the whole data center.
• Redundancy and Fault Tolerance:
• Multiple paths exist between any two servers.
• Ensures alternate routes in case of link failure.
• Failure of:
• Edge switch: Affects only a few servers.
• Aggregation/Core switch: Does not affect overall connectivity.
• Each pod contains:
• Edge switches
• Aggregation switches
• Server nodes (leaf nodes)
• Benefits:
• High bandwidth.
• Scalability for cloud applications with massive data transfer.
• Fault-tolerant due to multiple paths.
• Uses low-cost Ethernet switches (cost-effective).
• Routing handled internally by switches.
• Server nodes remain unaffected during switch failures (unless all paths fail).

Modular Data Center in Shipping Containers

1. Structure & Design:

• Modular data centers are housed in truck-towable containers.

• Inside each container:
• Racks filled with hundreds of blade servers.
• Racks are arranged around container walls.

2. Cooling Mechanism:

17
 • Fans and heat exchangers cool the hot air from servers.
• Chilled air or cold water circulates in a loop to maintain optimal temperature.
• Advanced cooling technology helps reduce cooling costs by up to 80%.

3. Capacity:

• Each SGI ICE Cube container can host:

• Up to 46,080 processing cores, or
• 30 petabytes (PB) of storage.

4. Benefits:

• High density of computing resources in a small space.

• Lower power consumption.
• Easy mobility – can be relocated to areas with:
• Cheaper electricity.
• Better water supply for cooling.
• Lower housing and maintenance costs.
• Faster deployment – containers can be assembled and tested quickly if resources
are ready.

Deployment Process:

• Starts with single server → moves to rack → expands to full container system.
• Building 40-server rack: ~half a day.
• Full container (1000+ servers): requires space planning, power, networking, and
cooling setup.
• Containers must be weatherproof and easy to transport.
• Suitable for cloud applications, like:
• Healthcare clinics needing local data centers.

 Interconnection of Modular Data Centers

Container-based data-center modules are meant for construction of even larger data
centers using a farm of container modules.

Ex:A Server-Centric Network for a Modular Data Center

18
 • Servers are shown as circles (e.g., 00, 01...).
• Switches are shown as rectangles (e.g., <0,0>, <1,0>...).
• Two levels in this example:
• Level 0: Basic server-to-switch connections (BCube0).
• Level 1: Higher-level connections that link different BCube0 groups.
• BCube0: Each switch connects to multiple servers (e.g., <0,0> connects to 00, 01, 02,
03).
• BCube1: Built from multiple BCube0 groups, connected using extra switches in Level
1.
• Servers have multiple ports.
• This allows each server to connect to:
• One Level 0 switch
• One Level 1 switch
• More ports → more redundancy and bandwidth.
• Many paths exist between any two servers.
• Ensures:
• High availability
• Fault tolerance
• Load balancing
• Better bandwidth utilization

 Inter-Module Connection Networks

Each container (like Container-00, Container-01, etc.) has a BCube network inside it
connecting servers and switches.

19
 • MDCube = BCube + Cube:
MDCube combines:
• A BCube structure inside the container
• A cube-like (grid) network among containers
• 2D Grid Example:
In the figure, 9 BCube1 containers (3x3 layout) are connected to form a 2D MDCube.
• Servers inside each container can communicate with other containers via high-speed
links between switches.
You can build large data centers by just adding more containers to the grid.
MDCube supports large cloud-based applications by providing high bandwidth,
fault-tolerant paths, and modular design.

 Data-Center Management Issues

 Making common users happy The data center should be designed to provide quality
service to the majority of users for at least 30 years
 Controlled information flow Information flow should be streamlined. Sustained
services and high availability (HA) are the primary goals.
 Multiuser manageability The system must be managed to support all functions of a
data center, including traffic flow, database updating, and server maintenance.
 Scalability to prepare for database growth The system should allow growth as
workload increases. The storage, processing, I/O, power, and cooling subsystems
should be scalable.
 Reliability in virtualized infrastructure Failover, fault tolerance, and VM live migration
should be integrated to enable recovery of critical applications from failures or
disasters.
 Low cost to both users and providers The cost to users and providers of the cloud
system built over the data centers should be reduced, including all operational costs.
 Security enforcement and data protection Data privacy and security defense
mechanisms must be deployed to protect the data center against network attacks and
system interrupts and to maintain data integrity from user abuses or network attacks
 Green information technology Saving power consumption and upgrading energy
efficiency are in high demand when designing and operating current and future data
centers

4.3 ARCHITECTURAL DESIGN OF COMPUTE AND STORAGE CLOUDS

 A Generic Cloud Architecture Design

 Cloud Platform Design Goals

 Key Design Goals of Cloud Computing Platforms:

20
 1. Scalability
• Easily add more servers, storage, or bandwidth as demand grows.
2. Virtualization
• Support both physical and virtual machines for flexible resource use.
3. Efficiency
• Combine hardware + software for easy and optimized operations.
4. Reliability
• Keep data in multiple locations (e.g., 3 disks in different data centers) to
avoid data loss.

 Cloud Management:

• Handles user requests, finds resources, and provisions them.

• Must support virtual and physical machines.
• Needs strong security due to shared resources.

 Enabling Technologies for Clouds

These technologies play instrumental roles in making cloud computing a reality. Most of
these technologies are mature today to meet increasing demand. In the hardware area, the
rapid progress in multicore CPUs, memory chips, and disk arrays has made it possible to
build faster data centers with huge amounts of storage space. Resource virtualization
enables rapid cloud deployment and disaster recovery.

21
 Generic Cloud Architecture(Ex:Google drive->file upload)

 The Internet cloud is seen as a massive cluster of servers.

 Servers (either physical or virtual machines) are provisioned on demand for web
services or distributed applications.
 Provisioning tools dynamically allocate or remove servers, software, and databases
to deliver services.
 Cloud platforms require distributed storage systems and related services.
 Built on data centers owned/operated by third-party providers.
 Software as a Service (SaaS) model abstracts technologies from users
 Emphasis on a distributed file system to handle large-scale data.

 Includes Storage Area Networks (SANs), database systems, firewalls, and security
devices.
 Web service APIs allow developers to access and use cloud resources.
 Monitoring and metering tools track performance and resource usage.
 Cloud software must automate resource management and maintenance.

 It must detect server status (joining or leaving) and respond accordingly.

 Providers like Google and Microsoft operate global data centers.
 Data center locations are optimized for power and cooling efficiency, often near
hydroelectric power sources.

 Private clouds offer easier management.

 Public clouds provide easier accessibility.
 The trend is moving toward hybrid clouds, combining the best of both.
 Interacting across intranet and public Internet is necessary for many apps.

 Security is critical for the operation of all cloud types.

 There is a high trust requirement due to the massive data involved.
 Focus on security and privacy is essential in future cloud studies.

22
  Layered Cloud Architectural Development (ex:Netflix Streaming Service)

• Cloud architecture is built in three main layers:

🔸 Infrastructure Layer (IaaS)
🔸 Platform Layer (PaaS)
🔸 Application Layer (SaaS)
• These layers are supported by virtualization and standardization of
hardware/software resources.
• Services are delivered to public, private, and hybrid clouds through
Internet/intranet connections.

Infrastructure Layer (IaaS)

• Foundation layer of cloud architecture.

• Built using virtualized compute, storage, and network resources.
• Offers flexibility, resource abstraction, and automated provisioning.
• Supports services like Amazon EC2.
• Demands minimal management effort from providers.
• Example: Amazon EC2, which offers virtual CPU and resource management.

Platform Layer (PaaS)

• Acts as middleware between infrastructure and application layers.

• Provides tools for:
• Application development
• Testing and operation flow monitoring
• Ensuring scalability, security, and reliability
• Offers general-purpose reusable software environments.
• Requires moderate management effort from providers.

23
Application Layer (SaaS)

• Topmost layer delivering software as a service to end-users.

• Includes apps for:
• Office work (document processing, calendar, info retrieval)
• Business use (CRM, sales, finance, supply chain)
• Used heavily in enterprise environments.
• Requires maximum provider effort – includes hardware, platform, and software.

Layer Dependency & Flexibility

• The layers are built from bottom-up:

• Infrastructure → Platform → Application
• Some services may span multiple layers depending on the needs.
• Example: Salesforce CRM covers all three layers.

Cloud Deployment Types

• Public Cloud: Accessed over the Internet (easy access).

• Private Cloud: Within intranet (easier to manage securely).
• Hybrid Cloud: Combines both; supports broader application use

 Market-Oriented Cloud Architecture

. Objective

Ensures QoS (Quality of Service) based on SLA (Service Level Agreement) between cloud
providers and users.

• Adopts a market-oriented approach instead of traditional resource management to:

• Handle supply and demand.
• Achieve market equilibrium.
• Promote economic incentives and cost efficiency.

24
Users/Brokers

• Users or brokers submit service requests to the cloud.

• They act as clients interacting with the data center/cloud infrastructure remotely.

SLA Resource Allocator

• Acts as an interface between users and cloud infrastructure.

• Manages resources according to QoS requirements.
• Involves several components:
• Service Request Examiner and Admission Control
• Pricing
• Accounting
• Service Request Monitor

Service Request Examiner and Admission Control

• Core logic for interpreting service requests.

• Handles:
• Customer-driven service management
• Computational risk management
• Autonomic resource management
• Decides to accept/reject requests based on:
• Current resource availability (from VM Monitor)
• Workload status (from Service Request Monitor)
• Assigns accepted requests to VMs and allocates resources accordingly.

Pricing Mechanism

• Determines how users are charged for their requests:

• Based on time (peak/off-peak)
• Based on pricing model (fixed/dynamic)
• Based on resource availability (demand/supply)
• Encourages efficient resource usage and user prioritization.

Accounting Mechanism

• Keeps records of actual resource usage.

• Helps in:
• Billing users.
• Providing usage history to enhance future allocation decisions.

VM Monitor

• Monitors the status and availability of Virtual Machines.

25
 • Tracks resource entitlements for each VM.
• Supports flexible VM allocation across physical machines.

Dispatcher

• Launches execution of service requests on assigned VMs.

• Coordinates with VM Monitor and Resource Allocator.

Service Request Monitor

• Tracks execution progress of service requests.

• Ensures efficient execution and informs the examiner for better future decisions.

Virtual Machines (VMs) and Physical Machines

• Physical machines host multiple VMs.

• Each VM can run isolated applications using different OS environments.
• VMs are started/stopped dynamically to handle user demand efficiently.

 Virtualization Support and Disaster Recovery

• Virtualization allows multiple Virtual Machines (VMs) to run on shared physical
servers.
• VMs act as containers for cloud services, enabling flexible and scalable deployments.
• Provisioning tools select physical servers and deploy VMs before running services.
• Cloud users and developers are abstracted from physical infrastructure details.
• Developers focus on application logic, not infrastructure concerns like scalability or
fault tolerance.
• Virtualization supports disaster recovery by allowing quick service migration to other
servers.
• Enhances resource utilization, fault tolerance, and service availability.

 Hardware virtualization

• Hardware virtualization uses specialized software to simulate hardware and run

unmodified operating systems.
• It allows legacy systems and unusual applications to run in cloud environments.
• Developers can use any OS or programming environment, ensuring consistency
between development and deployment.
• Virtualization software enables flexibility, allowing users to customize their system
environments.
• VMs (Virtual Machines) are commonly used to host third-party applications in the
cloud.
• VMs offer isolation, meaning users have full access to their own VMs without
affecting others.
• Multiple VMs with different OSes can run on the same physical server.

26
• Virtualization provides customized and separate environments for each user,
improving security and usability.
• Virtual disk storage and virtual networks support VM functionality.
• Virtualized resources (CPU, storage, etc.) are pooled and managed by virtual
integration managers.
• These managers handle load balancing, resource allocation, security, data, and
provisioning.

27
  Virtualization Support in Public Clouds

• AWS (Amazon Web Services): Offers full VM-level virtualization, allowing users to
run custom applications.
• Microsoft Azure: Offers programming-level virtualization through the .NET
framework.
• Google App Engine (GAE): Provides limited application-level virtualization with only
Google-managed services.
• Tools:
• VMware: Used for workstations, servers, and virtual infrastructures.
• Microsoft Tools: Used on PCs and special servers.
• XenEnterprise: Only for Xen-based servers.
• Benefits: Virtualization enables high availability (HA), disaster recovery, dynamic
load balancing, and rich provisioning, key for cloud and utility computing.

 Virtualization for IaaS (Infrastructure as a Service)

• VMs allow custom environments over shared infrastructure.

• Benefits of using VMs in IaaS:
• Server consolidation for better resource utilization.
• Legacy application support without compatibility issues.
• Security via sandboxing for untrusted applications.
• Performance isolation to ensure better Quality of Service (QoS) for users.

 Storage Virtualization for Green Data Centers

• IT power usage has doubled, contributing to ~3% of total US energy consumption.

• Fortune 500 companies are adopting energy-saving policies.
• Virtualization reduces power consumption and cuts costs by minimizing hardware
usage.
• Green data centers and storage virtualization further support energy-efficient
computing.

 VM Cloning for Disaster Recovery

 Disaster recovery ensures system availability during failures.

 Two recovery schemes:
o Physical-to-Physical: One physical machine recovers another.
o VM-to-VM: One virtual machine recovers another.
 VM recovery is faster, simpler, and cheaper than physical recovery.

 Traditional Physical Machine Recovery

 Involves:
o Hardware configuration
o Operating system installation

28
 o Backup agent setup
o Restarting system
 Results in long recovery time, often complex and expensive.

 VM-Based Recovery

 VM recovery skips several steps:

o No need to reinstall OS or backup agents
 Recovery time is reduced by ~60% compared to physical recovery.
 Enabled by VM encapsulation, where the entire VM state is packaged into a file.

 VM Cloning for Disaster Recovery

 Cloning involves making a replica VM on a remote server.

 Only one VM is active; the clone remains in suspended mode.
 In case of failure, the cloud control center activates the clone.
 Live migration occurs by sending:
o Snapshot of current state
o Only changed data to reduce downtime.

 Key Metrics

 Recovery Point Objective (RPO): Maximum acceptable data loss (determined by

how frequently snapshots are taken).
 Recovery Time Objective (RTO): Maximum acceptable downtime (reduced with fast
VM activation and migration).

 Security Consideration

 During live migration, VM security must be enforced to protect data and system
integrity.

4.3 ARCHITECTURAL DESIGN CHALLENGES

Challenge 1: Service Availability and Data Lock-in Problem

• Single-company-managed cloud services create single points of failure.

• Use of multiple cloud providers can improve availability.
• Threats like DDoS attacks can cut off SaaS income; rapid scale-ups can help.
• Proprietary APIs prevent easy migration of data/programs between clouds.
• Solution: Standardize APIs for interoperability and surge computing (dynamic load
sharing between private and public clouds).

Challenge 2: Data Privacy and Security Concerns

• Public cloud networks are more exposed to attacks.

• Security technologies like encryption, VLANs, firewalls can help.
• Legal requirement: Some countries mandate data localization.

29
 • Threats include traditional (DoS, malware) and cloud-specific (VM rootkits,
hypervisor attacks, man-in-the-middle attacks during migration).
• Need to secure both passive (data theft) and active (data manipulation) attacks.

Challenge 3: Unpredictable Performance and Bottlenecks

• VMs share CPU/memory well, but I/O sharing causes interference.

• Example: EC2 instance disk write speed significantly lower than expected.
• Data-intensive applications require careful placement and traffic control.
• Solution: Improve I/O architecture, widen bottleneck links, remove
underperforming servers.

Challenge 4: Distributed Storage and Widespread Software Bugs

• Need for scalable distributed storage (SAN) for growing databases.

• Difficult to reproduce and fix bugs at cloud scale in production.
• Solution: Use VMs for better debugging and traceability, and develop realistic
simulators for large-scale debugging.

Challenge 5: Cloud Scalability, Interoperability, and Standardization

• Scalability models differ (GAE auto-scales, AWS charges per VM-hour).

• Opportunity: Fast scalability without breaching SLAs.
• Use of Open Virtualization Format (OVF) for:
• Platform-independent VM packaging.
• Support for multi-VM appliances.
• Transport and integrity across virtualization platforms.
• Push for hypervisor-agnostic VMs and cross-platform live migration (e.g., Intel ↔
AMD).

Challenge 6: Software Licensing and Reputation Sharing

• Cloud adoption limited by traditional software licensing models.

• Open-source software popular; commercial licenses must adapt.
• Propose pay-per-use or bulk-use licensing schemes.
• Reputation issues: Misuse by one client (e.g., spamming) can blacklist the entire
cloud provider.
• Suggest reputation-guarding services like trusted email systems.
• Legal concerns over liability must be addressed in Service Level Agreements (SLAs).

4.4 PUBLIC CLOUD PLATFORMS: GAE, AWS, AND AZURE

 Public Clouds and Service Offerings

 Cloud Ecosystem Structure – 5 Levels of Cloud Players

30
1. Top Level – Users

 Individual Users: Primarily use applications provided via SaaS (Software as a

Service).
 Organizational Users: Typically utilize IaaS (Infrastructure as a Service) and PaaS
(Platform as a Service) for broader and more customizable computing needs.

2. SaaS Providers (Application Providers)

 Deliver ready-to-use applications over the internet.

 Cater mostly to individual users for tasks like email, document editing, CRM, etc.
 Depend on PaaS and IaaS layers to function.

3. PaaS Providers (Cloud Platform Providers)

 Offer a development and deployment environment in the cloud.

 Used by developers and enterprises to build, run, and manage applications.
 Support both organizational users and IaaS providers.

4. IaaS Providers (Cloud Service Providers)

 Deliver basic computing resources (compute power, storage, networking).

 Foundation for running SaaS and PaaS services.
 Serve both application providers and end-users (mainly organizations).

5. Base Infrastructure – Hardware & Software Providers

 Hardware Providers: Supply physical servers, storage systems, and networking gear.
 Software Providers: Offer system software, virtualization platforms, and
middleware.
 Together, they support the IaaS layer and indirectly all upper layers.

31
  Google App Engine (GAE)

Google Cloud Infrastructure

🔹 Massive Data Processing & Search Engine Expertise

 Google operates the world’s largest search engine, leveraging massive data
processing capabilities.
 This has led to innovations in data center design and scalable programming models
like MapReduce.

🔹 Global Data Center Network

 Google owns hundreds of data centers with over 460,000 servers worldwide.
 Up to 200 data centers can be used simultaneously for cloud applications.
 Data storage includes text, images, and video, with replication for fault tolerance
and high availability (HA).

Google App Engine (GAE) – PaaS Platform

 GAE is Google’s Platform as a Service (PaaS) launched in 2008.

 Supports cloud and web applications with scalability (elasticity).
 Enables users to deploy and run applications on Google’s global infrastructure.
 GAE is becoming a popular platform for small cloud service providers.

Key Technologies Behind Google Cloud

 Google File System (GFS) – Distributed file storage.

 MapReduce – Parallel processing of large datasets.
 BigTable – Distributed database system.
 Chubby – Lock service for managing distributed systems.

32
  GAE Architecture

1. Google File System (GFS)

 A distributed file system designed to manage large volumes of data across multiple
machines.
 GFS Master: Manages metadata (file names, locations).
 GFS Chunkservers: Store actual data chunks and serve them upon request.
 Ensures fault tolerance through data replication.

2. MapReduce

 A programming model and processing engine for large-scale data processing.

 Map function processes key-value input pairs and generates intermediate results.
 Reduce function aggregates the intermediate results.
 Supports parallel processing across thousands of nodes.

3. BigTable

 A highly scalable, distributed database system.

 Used for structured and semi-structured data.
 Provides fast read/write access and is suitable for big data analytics.
 Forms the foundation of services like Google Search, Gmail, and Google Analytics.

4. Chubby

 A lock service for managing distributed systems.

 Ensures that different services (like GFS or BigTable) coordinate access to shared
resources safely.
 Helps in leader election, synchronization, and metadata management.

5. Scheduler

 Assigns tasks (e.g., MapReduce jobs) to different nodes (servers).

 Ensures efficient resource utilization, load balancing, and job prioritization.

33
  User sends requests through an application interface (like a browser or app).
 The request enters the Google cloud infrastructure (shaded cloud area).
 Inside the infrastructure:
o Scheduler allocates resources.
o Chubby coordinates access.
o GFS Master controls file system metadata.
o Multiple nodes run services like:
 MapReduce Jobs
 BigTable Server
 GFS chunkserver
 Scheduler slave
 All run over Linux OS.

 Functional Modules of GAE

The GAE platform comprises the following five major components. The GAE is not an
infrastructure platform, but rather an application development platform for users. We
describe the component functionalities separately.

a. The datastore offers object-oriented, distributed, structured data storage services based
onBigTable techniques. The datastore secures data management operations.

b. The application runtime environment offers a platform for scalable web programming
andexecution. It supports two development languages: Python and Java.

c. The software development kit (SDK) is used for local application development. The SDK
allowsusers to execute test runs of local applications and upload application code.

d. The administration console is used for easy management of user application

development cycles,instead of for physical resource management.

e. The GAE web service infrastructure provides special interfaces to guarantee flexible use
and management of storage and network resources by GAE.

 GAE Applications
 Well-known GAE applications include the Google Search Engine, Google Docs,
Google Earth, and Gmail. These applications can support large numbers of users
simultaneously.
 Users can interact with Google applications via the web interface provided by each
application. Third-party application providers can use GAE to build cloud applications
for providing services.
 The applications are all run in the Google data centers. Inside each data center, there
might be thousands of server nodes to form different clusters. Each cluster can run
multipurpose servers.
 GAE supports many web applications. One is a storage service to store application-
specific data in the Google infrastructure. The data can be persistently stored in the

34
 backend storage server while still providing the facility for queries, sorting, and even
transactions similar to traditional database systems.
 GAE also provides Google-specific services, such as the Gmail account service (which
is the login service, that is, applications can use the Gmail account directly). This can
eliminate the tedious work of building customized user management components in
web applications. Thus, web applications built on top of GAE can use the APIs
authenticating users and sending e-mail using Google accounts.

 AWS

 Amazon Web Services (AWS) is a public cloud computing platform using the
Infrastructure-as-a-Service (IaaS) model.
 It allows developers and companies to build and run applications in the cloud using
virtual machines and other cloud-based resources.

 AWS Architecture

1. EC2 Instances (Elastic Compute Cloud)

 Acts like virtual servers to run applications.

 Shown in the image as multiple black servers
labeled "EC2".

2. EBS (Elastic Block Store)

 Provides persistent block-level storage to EC2 instances

(attached like a hard drive).

3. S3 (Simple Storage Service)

o Used for storing objects like files and images.
o Acts as a central storage point.
4. SQS (Simple Queue Service)
o Enables reliable message delivery between different services or components
(e.g., EC2 to EC2).
5. SimpleDB
o A non-relational database service used to store and query structured data.
6. Developer Interface
o Developers use tools to interact with the cloud resources (through APIs, SDKs,
and management consoles).
7. User Interaction
o Users interact with applications running on EC2 through a web browser or app.

Additional Features

 CloudWatch: Monitors AWS resources like EC2 for CPU, disk, and network metrics.

35
  Elastic Load Balancer (ELB): Distributes traffic across multiple EC2s for high
availability.
 Auto Scaling: Automatically adjusts EC2 count based on load.
 Amazon DevPay & FPS: Handle billing and payment integration for commercial cloud
services.
 Amazon Mechanical Turk: Offers a workforce-as-a-service model for tasks like data
labeling or surveys.
 MPI Clusters (since 2010): High-performance computing using cluster compute
instances.

 Microsoft Windows Azure

 Launched by Microsoft in 2008 to meet cloud computing challenges.

 Built over Microsoft-managed data centers.
 Based on Windows OS and Microsoft virtualization technology.
 Applications are deployed on Virtual Machines (VMs) within data centers.

36
  Architecture

 Divided into three major component platforms.

 Manages servers, storage, and networking infrastructure.
 Sits beneath a layer of cloud-level services used to build various cloud applications.

 Cloud-Level Services in Azure

1. Live Service
o Access to Microsoft Live applications.
o Supports data usage across multiple machines concurrently.
2. .NET Service
o Enables application development on local hosts.
o Applications can run on cloud-based machines.
3. SQL Azure
o Provides access to relational database services using SQL Server in the cloud.
4. SharePoint Service
o A platform for developing custom business web applications.
o Scalable and manageable.
5. Dynamic CRM Service
o Supports business applications related to finance, marketing, sales, and
promotions.

 Integration and Interoperability

 Compatible with traditional Microsoft software:

o Windows Live, Office Live, Exchange Online, SharePoint Online, Dynamic CRM
Online.
 Uses standard web communication protocols:
o SOAP and REST.
 Supports integration with other platforms and third-party clouds.

 Azure Development Kit (SDK) available for download.

 Enables:
 Local development and testing of Azure applications.
 Debugging on Windows host systems.

4.5 INTER-CLOUD RESOURCE MANAGEMENT

37
 Six Layers of Cloud Services Architecture

Bottom 3 – Physical Layers (Resource Foundation)

1. HaaS (Hardware as a Service)

o Physical hardware resources are provided as services.
2. NaaS (Network as a Service)
o Provides connectivity between hardware (e.g., Virtual LANs).
3. LaaS (Location as a Service)
o Also called collocation services.
o Offers physical housing, power, and security for hardware and network.
o Sometimes referred to as Security as a Service.

🔼 Top 3 – Service Layers (Cloud Service Models)

4. IaaS (Infrastructure as a Service)

o Provides compute, storage, Data as a Service (DaaS), and Communication as
a Service (CaaS).
5. PaaS (Platform as a Service)
o Built on top of IaaS.
o Offers a platform for developing, testing, and deploying applications.
6. SaaS (Software as a Service)
o Topmost layer.
o Offers fully functional applications over the cloud.

⚠️ Dependency: Each upper layer depends on the layer below it.

Example: SaaS cannot function without PaaS → PaaS needs IaaS → IaaS needs
HaaS/NaaS/LaaS.

 Cloud Player Categories

1. Cloud Providers / IT Administrators

o Focus on infrastructure performance (IaaS layer).
2. Software Developers / Vendors
o Focus on application performance on the platform (PaaS/SaaS layers).
3. End Users / Business Users
o Focus on quality of service (QoS) and security (mainly SaaS).

38
 Cloud Service Tasks and Trends

1. SaaS (Software as a Service)

 Top cloud layer, focused on business applications.

 Key areas: CRM, HR, finance, collaboration.
 CRM: First successful SaaS, used for marketing, sales, customer analytics.
 Rapid growth in adoption across industries.

2. PaaS (Platform as a Service)

 Enables application development and deployment.

 Major providers: Google, Salesforce, Facebook.

3. IaaS (Infrastructure as a Service)

 Provides virtualized computing resources.

 Major providers: Amazon, Windows Azure, RackRack.

4. Collocation Services

 Multiple cloud providers collaborate.

 Used in manufacturing for supply chain support.

5. Network Cloud Services

 Focused on communication infrastructure.

 Examples: AT&T, Qwest, AboveNet.

6. Vertical Cloud Services

 Layered, interdependent services.

 Often involve cloud mashups for integrated applications.

 Software Stack for Cloud Computing

 Purpose and Requirements

 Cloud software stacks are custom-built to ensure:

o High throughput
o High availability (HA)
o Fault tolerance
 OS may be modified to meet cloud-specific needs.

39
  Layered Software Stack Structure

 Inspired by major providers (e.g., Google, Microsoft, Yahoo!).

 Stack is layered, with each layer:
o Serving a specific purpose.
o Providing interfaces to the upper layers.
 Unlike traditional stacks, lower layers aren’t fully transparent to upper ones.

 Platform Layer

 Can run on physical or virtual servers.

 Use of VMs enables platform flexibility and hardware independence.

 Data Storage Layer

 Sits on top of the platform.

 Acts like a file system, managing massive data storage.

 Application Execution Layers

 Run on top of the storage layer.

 Includes:
o Database storage systems
o Large-scale cluster programming
o Data query language support

 Software Stack Components

 Upper layers contain key cloud application components and services.

 Each component builds on the services provided by the layer beneath it.

 Runtime Support Services

 Similar to cluster environments, cloud systems require runtime services for smooth
operation.
 Ensures efficiency and proper functioning of the cloud cluster.

 Cluster Monitoring

 Gathers real-time status of the cloud cluster.

 Enables performance tracking and issue detection.

 Job Management and Scheduling

 Tasks are queued and assigned to nodes based on availability.

 Uses a distributed scheduler, often optimized for MapReduce-style applications.

40
 SaaS Runtime Environment

 Supports browser-based applications for thousands of users.

 No need for customers to invest in hardware or software licenses.
 Reduces provider costs vs. traditional hosting models.

 Data Storage

 Customer data resides in:

o Vendor-proprietary clouds or
o Public clouds supporting PaaS and IaaS models.

41