Message Passing, Remote Procedure Calls and Distributed Shared Memory as Communication Paradigms for Distributed Systems & Remote Procedure Call Implementation Using Distributed Algorithms

1. Advance Operating
System
1
Presented By:
Zunera Altaf
Sehrish Asif
Wajeeha Malik

2. Reference:
Message Passing, Remote Procedure Calls and
Distributed Shared Memory as Communication Paradigms for Distributed Systems
&&
Remote Procedure Call Implementation Using Distributed Algorithms
Given By:
J. Silcock and A. Goscinski
{jackie, ang@deakin.edu.au}
School of Computing and Mathematics
Deakin University
Geelong, Australia
&&
G. MURALI i K.ANUSHAii A. SHIRISHAiii S.SRAVYAiv Assistant Professor,
Dept of Computer Science Engineering
JNTU-Pulivendula, AP, India
2

3. MessagePassing-MP
RemoteProcedureCall-RPC
DistributedSharedMemory-DSM
3
Contents:

4. Introduction
• Distributed System
• Distributed Programming
• Reasons for using DS
• To connect several computers first application that may require to the use of
communication network to communicate.
• Second reason to use distributed system is single computer may be possible in principle
• Examples of distributed systems
• Telephone networks and cellular networks
• Computer networks such as the Internet
• Wireless sensor network.
• Applications of distributed computing
• World wide web and peer-to-peer networks.
• Massively multiplayer online games and virtual reality communities
• Distributed databases and distributed database management systems 4

5. Message Passing - Introduction
• It requires the programmer to know
Message
Name of source
Destination process.
Sender message passing Receiver
5
Send(recieve,
msg, type)
Send(recieve,ms
g,type)

6. Message Passing
• Message passing is the basis of most interprocess
communication in distributed systems. It is at the lowest level
of abstraction and requires the application programmer to be
able to identify
 Message.
 Name of source.
 Destination process.
 Data types expected for process.
6

7. Syntax of MP
• Communication in the message passing paradigm, in its
simplest form, is performed using the send() and receive()
primitives.
• The syntax is generally of the form:
• send(receiver, message)
• receive(sender, message)
7

8. Semantics of MP
• Decisions have to be made, at the operating system level,
regarding the semantics of the send() and receive() primitives.
• The most fundamental of these are the choices between
blocking and non-blocking primitives and reliable and
unreliable primitives.
8

9. Semantics(cont.)
• Blocking: Blocking will wait until a communication has been completed in its
local process before continuing.
• A blocking send operation will not return until the message has entered into MP internal
buffer to be completed.
• A blocking receive operation will wait until a message has been received and completely
decoded before returning.
• Non-Blocking
• Non-blocking will initiate a communication without waiting for that communication to be
completed.
• The call to non-blocking operation(send/ receive) will return as soon as the operation is
began, not when it completes
• Pros & cons of non-blocking: This has the advantage of not leaving the CPU idle while the
send is being completed. However, the disadvantage of this approach is that the sender does
not know and will not be informed when the message buffer has been cleared. 9

10. Semantics(cont.)
• Buffering
• A unbuffered send & receive operation means that the sending
process sends the message directly to the receiving process rather than a
message buffer. The address, receiver, in the send() is the address of the
process and same as recieve()
• There is a problem, in the unbuffered case, if the send() is called before
the receive() because the address in the send does not refer to any
existing process on the server machine.
sender process receiver process
10

11. Semantics (Cont…)
• A buffered send & receive
• operation will send address of buffer in send() and recieve() as
destination parameter to communicate with other process.
• Buffered messages save in buffer until the server process is ready to
process them.
11
Sender: Send
(buff,type,msg)
Receiver
buffer

12. Semantics (Cont.) - Reliability
Fig: A Unreliable Send
Fig: A Reliable Send
12
Sender Receiver
Sender process waiting for
response without knowing
message is not transmitted to
receiver
Sender Receiver
Message received

13. Semantics(cont…)
• Direct/indirect communication: Ports allow indirect
communication. Messages are sent to the port by the sender and
received from the port by the receiver. Direct communication
involves the message being sent direct to the process itself, which is
named explicitly in the send, rather than to the intermediate port.
• Fixed/variable size messages: Fixed size messages have their size
restricted by the system. The implementation of variable size
messages is more difficult but makes programming easier, the
reverse is true for fixed size messages.
13

14. Message passing Demo code
class Program
{
static void Main(string[] args)
{
using (new MPI.Environment(ref args))
{
ntracommunicator comm = Communicator.world;
int number = 6;
if (comm.Rank == 0)
{
for (int i = 1; i < comm.Size&& number>0; i++ ,number--
)
{
comm.Send<int>(number, i, 0);
int[] ansew =comm.Receive<int[]>(i, 1);
for (int j = 0; j < ansew.Length; j++)
{
Console.WriteLine(number+ " * "+ j+ " = "+ansew[j]);
}
}
}
else
{
int size = 10;
int[] array = new int[size];
int X = comm.Receive<int>(0, 0);
for (int i = 0; i < size; i++) {
int product = X * i;
array[i]= product;
}
comm.Send<int[]>(array, 0, 1);
Console.WriteLine("rank :" + comm.Rank);
}
}
}
}
14

15. Output
Rank 1 Rank 2
1*1=1 4*1=4
1*2=2 4*2=8
1*3=3 4*3=12
1*4=4 4*4=16
1*5=5 4*5=20
1*6=6 4*6=24
1*7=7 4*7=28
1*8=8 4*8=32
1*9=9 4*9=36
1*10=10 4*10=40
Rank 4 Rank 5
2*1=2 3*1=3
2*2=4 3*2=6
2*3=6 3*3=9
2*4=8 3*4=12
2*5=10 3*5=15
2*6=12 3*6=18
2*7=14 3*7=21
2*8=16 3*8=24
2*9=18 3*9=27
2*10=20 3*10=40
15

16. Producer consumer example by MP
16

17. Implementation Screen Shots
17

18. Message
Passing
• Explicit control of the
movement of data
RPC
• Level of
abstraction
increases
18

19. RPC: Remote Procedure Calling
19
For client-server based applications RPC is a
powerful technique between distributed
processes
Calling procedure, Called procedure need not
exist in the same address space.
Two systems may be on same system or they
may be on the different systems.

20. Syntax
20
• call
procedure_name(value_arguments;
result_arguments)
Call
• receive procedure_name(in
value_parameters; out
result_parameters)
Receive
• reply(caller, result_parameters)Reply

21. Semantics: How RPC Works
21

22. Producer Consumer Example Using
RPC
22

23. TSP: Travelling Salesman Problem
• The traditional lines of attack for the NP-hard problems are
the following:
• 1. For finding exact solutions:
• Fast only for small problem sizes.
• 2. Devising suboptimal algorithms:
• Probably provides good solution
• Not proved to be optimal.
• 3. Finding special cases for the problem:
• Either better or exact heuristics are possible.
• Given a set of cities and the distance between each possible pair,
the Travelling Salesman Problem is to find the best possible way of
‘visiting all the cities exactly once and returning to the starting
point’. 23

24. TSP: Travelling Salesman Problem
24

25. Using Parallel Branch & Bound
• B&B uses a tree for TSP.
25
Node of the tree represents
a partial tour for the use of
TSP.
Leaf
represents a
solution of
problem.

26. Issuesregardingthepropertiesofremoteprocedurecallstransparency:
Binding
26
• (NameLocation)
• Implemented at the operating system
level, using a static or dynamic linker
extension.
• Another method is to use procedure
variables which contain a value which is
linked to the procedure location.
Communication and site failures can result in
inconsistent data because of partially completed
processes. The solution to this problem is often left to
the application programmer.
Parameter passing in most systems is restricted to
the use of value parameters.
Exception handling is a problem also associated with
heterogeneity. The exceptions available in different
languages vary and have to be limited to the lowest
common denominator.
Communication Transparency: The user should be unaware that the
procedure they are calling is remote.

27. Should not interfere with communication
mechanisms.
Single threaded clients and servers, when blocked
while waiting for the results from a RPC, can cause
significant delays.
Lightweight processes allow the server to execute calls
from more than one client concurrently.
ConcurrencyHeterogeneity
27
Machines may have different data representations
Machines may be running different operating system
Remote procedure may have been written using a different
language
Static interface declarations of remote procedure serve to establish
agreement between the communicating processes
Issues regarding the properties of remote procedure calls transparency:

28. • Memory which although distributed over a network of autonomous
computers and is accessed through Virtual addresses.
• DSM allows programmers to use shared memory style programming.
• Programmers are able to access complex data structures.
• OS has to send messages b/w machines with request for memory not
available locally make replicated memory consistent
Distributed Shared Memory
28

29. Syntax
29
The syntax used for DSM is the same as that of normal centralized memory
multiprocessor systems.
-> read(shared_variable)
-> write(data, shared_variable)
• read() primitive requires the name of the variable to be read as its argument.
• write() primitive requires the data and the name of the variable to which the data is
to be written.
variable through its virtual address

30. Semantics:ThereareseveralissuesrelatedtothesemanticsofDSM:
1.Structure&Granularity:
Structure:Memorycantakeunstructuredlineararrayofwords orstructuredformsof
objects.
Granularity:relatestothesizeofthechunksoftheshareddata.
Coarse Grained Solution
• It is a page-based distributed memory management
• Is an attempt to implement a virtual memory model where paging
takes place over a network instead of to disk
Finer Grained Solution
• It can lead to higher network traffic
30

31. Semantics(cont)
2.Consistency
31
Consistency
Simplest implementation of shared
memory a request for a non local piece of
data
Is very similar to thrashing in virtual
memory and it leads to lowering
performance
Consistency models determine the
conditions :memory updates will be
propagated through the system

32. Semantics(cont)….
3.Synchronization: Shared data must protect by synchronization
primitives,locks,monitors or semaphores.
32
• It can be managed by synchronization manager
• It can be made the responsibility of the application
developer
• Finally it can be made the responsibility of the system
developer

33. Semantics(cont)….
4. Heterogeneity: Sharing data b/w heterogeneous machines is an
important problem.
4.1 Mermaid Approach:
• Is allow only one type of data on an appropriately tagged page
• The overhead of converting that data might be too high to make DSM on
heterogeneous system
5.Scalability: this is one of the benefits of DSM systems
• Is that they scale better than many tightly-coupled multiprocessors.
• Is limited by physical bottlenecks
33

34. ProducerConsumersupportedbyDSMatSystemLevel
Syntax of the code is
same as that for the
centralized version but
implications of the OS
are quite different.
The Process
requiring a lock on
semaphore
executes
wait(sem).
When process has
completed the
critical region it will
execute a signal
(sem)
34
Best method of measuring the value of the DSM is to measure its performance against
the other communication paradigms..

35. ProducerConsumersupportedbyDSMatUserLevel
It is based on the implementation.
In this implementation shared
memory exist at user level in the
memory space of the server.
The producer and consumer both
use remote procedure calls to call
the cmm.
35

36. Analysis:
Ease of implementation for
system user
Ease of use at application
programming level
Performance
36
This analysis of the implementations of the producer-consumer problem using
message passing, RPC and DSM at user and operating system level will be
based on the following three criteria:

37. EaseofimplementationforSystemuser:

37
When using message passing paradigm for
communication b/w processes the system
must provide mechanism for passing
messages b/w processes.
Code written at system level is complex
compared with message passing and RPC.
In the final ,DSM implemented at user level.
System designer must provide message
passing and RPC and then use these two
mechanisms to implement a central
memory manager.

38. Easeofuseatapplicationprogramminglevel:
In the case of message passing ,application
programmers must be aware of the semantics of
the implementation of send and receive
primitives.
Using RPC application programmers can call
procedure without knowing that it is remote and
can program.
DSM at system level is easiest paradigm to use at
programming level, provides a familiar programming
model.
DSM at user level is simple for the application
programmer to use as the producer produces item and
calls the central memory manager.
38

39. Performance:
Interprocess communication consumes a large amount of
time, a relative measure of performance b/w the three
paradigms
The message passing implementation requires only 1 message
to be passed while RPC requires 2messages while DSM
implemented at OS level requires 12 ages and at user level
requires 4 messages.
DSM has a large overhead which must be minimized as much as
possible
39

40. Conclusion..
• We have discussed three high level communication paradigms for
DOS.
• Message passing,RPC and DSM and base their discussion on their
semantics ,syntax and discuss the implications of implementation
and use of these paradigms for the application programmer and the
OS designer.
40

41. 41