Principles of Operating

Systems

Naresh Chauhan

© Oxford University Press 2014. All rights reserved.

 Chapter 18

Distributed Operating
Systems

© Oxford University Press 2014. All rights reserved.

 Objectives
• Characteristics of a distributed system
• Differentiation between network and distributed operating system
• Various issues in distributed operating system
• IPC methods in distributed operating system
• Concept of global clock and logical clock
• Lamport’s Logical clock algorithm for synchronization
• Global distributed system state
• Achieving global state through Chandy-Lamport algorithm
• Algorithms to resolve mutual exclusion problem
• Deadlock detection through centralized and distributed algorithms
• Deadlock prevention through wait-die and wound-wait algorithms
• Distributed process scheduling algorithms
• Distributed shared memory

© Oxford University Press 2014. All rights reserved.

 Introduction to distributed systems
• A distributed system is a loosely coupled architecture wherein
processors are inter-connected by a communication network.
• Distributed systems are multiprocessor systems but with the
following differences:
▪ Distributed system works in a wide area network involving much
more communication as compared to computation.
▪ Each node in a distributed system is a complete computer
having full set of peripherals including memory, communication
hardware, possibly different operating system and different file
system, etc.
▪ The users of a distributed system have an impression that they
are working on a single machine.

© Oxford University Press 2014. All rights reserved.

 Difference between network operating
system & distributed operating system
• Network operating system are loosely coupled operating system
software on a loosely coupled hardware that allows nodes and
users of a distributed system to be quite independent of one
another but interacts in a limited degree.
• Distributed operating systems are tightly coupled operating system
software on loosely coupled hardware, i.e. distributed system.
• Since in a distributed system, each node is a complete computer
system, there is no global memory and even no global clock.

© Oxford University Press 2014. All rights reserved.

Network Operating System

© Oxford University Press 2013. All rights reserved.

 Communication in distributed operating
system
• Inter Process Communication is achieved in Distributed Systems is
implemented in two ways:
▪ Message Passing Model
▪ Remote Procedure Call

© Oxford University Press 2014. All rights reserved.

Message Passing Protocol

© Oxford University Press 2013. All rights reserved.

Implementation of RPC

© Oxford University Press 2013. All rights reserved.

Clock synchronization in distributed system
Synchronizing Logical Clocks
• Lamport defined a relation known as Happens-before relation.
This relation is basically for capturing the underlying
dependencies between events. It may be denoted as x → y. There
may be following two situations for this relation:
• If x and y are events in the same process and x occurs before y.
• If x is the message sending event by one process and y is the
message receiving event. In this case x will always precede y.
• Happens-before relation is a transitive relation, i.e.,
• If x → y, y → z, then x → z.

© Oxford University Press 2014. All rights reserved.

Clock synchronization in distributed system
• In distributed systems, it is better to know the sequence of events
using this relation and by ascertaining the order of events help in
designing, debugging, and understanding the sequence of
execution in distributed computation. It is clear that when one
event changes the system state, it may affect its related future
events that will happen after this. This influence among
causally-related events satisfying the Happens-before relation, are
known as causal affects.

© Oxford University Press 2014. All rights reserved.

Clock synchronization in distributed system
• If two events are with the following conditions:
– x → y, i.e. x → y is not true.

– y → x, i.e. y → x is not true.

• There is no transitive Happens-before relationship between x and

y.
• Then these two events are known as concurrent, i.e. it cannot be
decided when of the two events happened or which of the two
first happens.

© Oxford University Press 2014. All rights reserved.

 Lamport’s Logical Clock
• To implement the logical clock, Lamport introduced the concept of
timestamp to be attached with each process in the system. The
timestamp may be a function that assigns a number Ti(x) to any
event x to process Pi. But these timestamps have nothing to do
with actual real time clock. Now as per the timestamp the
following conditions must be met:
• For any two events x and y in a process Pi, If x → y , then
T(x) < T(y)
• If x is the message sending event in process Pi and y is the message
receiving event in process Pj then
Ti(x) < Tj(y)

© Oxford University Press 2014. All rights reserved.

 Lamport’s Logical Clock
• The timestamp value T must always be increasing and never
decreasing. Thus, for the implementation, the clock is incremented
between any two successive events with a positive integer always,
i.e.,
Ti(y) = Ti(x) + d where d >0
• Using all the conditions mentioned above, we can assign
timestamp to all events in the system and thereby provide a total
ordering of all the events in the system.

© Oxford University Press 2014. All rights reserved.

 Global State
• In a distributed system, there is no global memory, a process if
wishes to know the state of an entire system, is not available to it.
• However, the up-to-date state of the full system is required for
system’s behavior, debugging, fault recovery, synchronization, etc.
• But in the absence of global clock and global memory, a coherent
global state of the system is difficult to achieve.
• Chandy Lamport have given an algorithm to achieve the same.

© Oxford University Press 2014. All rights reserved.

 Chandy-Lamport Consistent State
Recording Algorithm
Marker sending rule for a process Ps
1.Ps records its state.
2.for (i=1; i =n; ++i) {
If marker not sent to Ci
Ps sends a marker on each outgoing Ci
}
Marker receiving rule for a process Q
•Process Q receives the marker message through a Ci.
•If Q has not recorded its state earlier
Call Marker sending rule to record its state and
to send the marker to each channel
else
Record the state of the channel as sequence of
messages received after Q’s state was recorded
and before Q received the current marker.

© Oxford University Press 2013. All rights reserved.

 Mutual Exclusion
To provide mutual exclusion among processes in distributed system
the following algorithms are used:
▪ Centralized algorithm: In this algorithm, a process on one
node of the distributed system is assigned as coordinator to
manage the mutual exclusion problem.
▪ Ricart-Agarwala algorithm: It is a fully distributed algorithm.
According to this algorithm, a process wishing to enter its
critical section sends a time-stamped request messages to all
other processes and waits.
▪ Token-ring algorithm: This algorithm assumes a logical
structure of all the processes in a ring, i.e. all the processes are
connected together in a ring sequence and every process
knows who the next process in that sequence is.

© Oxford University Press 2014. All rights reserved.

 Richart-Agrawala Algorithm
Ricart-Agarwala Algorithm for Mutual Exclusion
1.A process Pi wishing to enter its critical section CS sends a message M(Name, Process_ID, TS) to all other
processes, i.e. n-1 in the system; where n is the number of processes in the system.
2.When a process Pj receives the message M from Pi, the following actions may be taken:
If Pi is not inside CS
Send OK message to Pi.
Else
{
If Pj has sent M to other processes
{
If (TSi < TSj) // Compare the time stamps of Pi and Pj.
Send OK message to Pi.
Else
Add the Pi ‘s request to Pending_request_queue.
}
Else
{
If Pj is inside the CS
Add the Pi ‘s request to Pending_request_queue.
}
}
1.The process if receives n-1 OK messages from the processes, it gets the permission to enter its CS and starts
executing inside it.
2. If the process Pj exits its CS
{
If items are there in Pending_request_queue
Send OK message to all the processes inside the
Pending_request_queue
}
© Oxford University Press 2013. All rights reserved.
Token-Ring algorithm for mutual exclusion

© Oxford University Press 2013. All rights reserved.

 DeadLock Detection
• To provide deadlock detection in distributed system the following
algorithms are used:
– Centralized algorithm: In this algorithm a process is assigned as
a central coordinator that will maintain the resource allocation
state of the entire system.
– Distributed algorithms: All the nodes in the system collectively
cooperate to detect a cycle.

© Oxford University Press 2014. All rights reserved.

 DeadLock Prevention
• To provide deadlock prevention in distributed system the following
algorithms are used:
– Wait-die algorithm: The older process waits for a younger
process but a younger process is killed if tries to wait for an
older process
– Wound-wait algorithm: The older process preempts the
younger one but if younger one requests, it is allowed to wait.

© Oxford University Press 2014. All rights reserved.

 Distributed Process Scheduling
• To perform distributed process scheduling, Load balancing is
performed by transferring some processes from heavily loaded
node to lightly loaded node. This is known as process migration.
• There are two types of schedulers.
▪ Local scheduler
▪ Global load scheduler
• Global load scheduler is of two types: static and dynamic. Static
scheduler assigns the processes to processors at their compile time
only while dynamic scheduler assigns the processors when the
processes start execution.

© Oxford University Press 2014. All rights reserved.

 Distributed file systems
• In distributed system, to have a distributed file system, some nodes
are dedicated to store the files only. These nodes perform the
storage and retrieval operation on the files. These nodes are known
as file servers. The other nodes used for computational purposes
are known as clients.

© Oxford University Press 2014. All rights reserved.