Interprocess Communication

Tanenbaum, van Steen: Ch2 (Ch3) CoDoKi: Ch2, Ch3, Ch5

Fall 2008 Jussi Kangasharju

Chapter Outline
Overview of interprocess communication Remote invocations (RPC etc.) Message passing Streams Publish/subscribe Multicast

Kangasharju: Distributed Systems

Middleware Protocols

General purpose services -Naming, browsing -Security -Atomicity -Higher-level communication -RPC, RMI -Message passing -Reliable multicast

An adapted reference model for networked communication.

Kangasharju: Distributed Systems

Remote Procedure Calls

Basic idea:
passive Available Executed

routines for remote clients by a local worker process, invoked by local infrastructure

See examples in book

Kangasharju: Distributed Systems

RPC goals
Achieve access transparent procedure call Cannot fully imitate
naming, global

failures, performance vs. call-by-value

variables, context dependent variables, pointers

Call-by-reference

Call semantics
Maybe,

at-least-once, at-most-once delivery RPC

Exception

Can be enhanced with other properties

Asynchronous Multicast, Location

broadcast processing
5

transparency, migration transparency,

Concurrent

Kangasharju: Distributed Systems

RPC: a Schematic View

System A X, Y, Y Z Thread P Y=FNCT(X,Y)

System B

FNCT(a,b) c:={comp} return c.

a:=X; b:=Y; RPC package

Kangasharju: Distributed Systems

RPC package
6

Implementation of RPC
RPC components:
RPC

Service (two stubs)

- interpretation of the service interface - packing of parameters for transportation

Transportation

service: node to node

- responsible for message passing - part of the operating system

Name service: look up, binding

name

of procedure, interface definition

Kangasharju: Distributed Systems

Passing Value Parameters

Steps involved in doing remote computation through RPC

Kangasharju: Distributed Systems

Writing a Client and a Server

The steps in writing a client and a server in DCE RPC.

Kangasharju: Distributed Systems

Binding a Client to a Server

Client-to-server binding in DCE.

Kangasharju: Distributed Systems

10

Implementation of RPC
Server: who will execute the procedure? One server process
infinite call one

loop, waiting in receive

arrives : the process starts to execute call at a time, no mutual exclusion problems possible

A process is created to execute the procedure

parallelism overhead mutual

exclusion problems to be solved

One process, a set of thread skeletons:

one

thread allocated for each call

Kangasharju: Distributed Systems

11

Distributed Objects
Remote Method Invocation ~ RPC A distributed interface
binding: server

download the interface to the client => proxy

stub ~ skeleton on a single machine (possible distribution: hidden)

The object
resides if

needed: object look through an adapter object may be persistent or transient system-wide

an

Object references:
typically: binding:

implicit or explicit resolving of an object reference

Binding and invocation Examples: CORBA, DCOM (Ch. 10)

Kangasharju: Distributed Systems 12

Distributed Objects

Fig. 2-16. Common organization of a remote object with client-side proxy.

Kangasharju: Distributed Systems

13

Object Adapter

Fig. 3-8. Organization of an object server supporting different activation policies.

Kangasharju: Distributed Systems

14

Binding a Client to an Object

Distr_object* obj_ref; obj_ref = ; obj_ref-> do_something(); (a) Distr_object objPref; Local_object* obj_ptr; obj_ref = ; obj_ptr = bind(obj_ref); obj_ptr -> do_something();
(b)
Fig. 2-17. (a) Example with implicit binding using only global references (b) Example with explicit binding using global and local references

//Declare a systemwide object reference // Initialize the reference to a distributed object // Implicitly bind and invoke a method //Declare a systemwide object reference //Declare a pointer to local objects //Initialize the reference to a distributed object //Explicitly bind and obtain a pointer to // the local proxy //Invoke a method on the local proxy

Kangasharju: Distributed Systems

15

Parameter Passing

Fig. 2-18.The situation when passing an object by reference or by value.

Kangasharju: Distributed Systems

16

Design Issues
Language independent interface definition Exception handling Delivery guarantees
RPC

/ RMI semantics

maybe at-least-once at-most-once (un-achievable:

exactly-once)

Transparency (algorithmic vs. behavioral)

Kangasharju: Distributed Systems

17

RPC: Types of failures

Client unable to locate server Request message lost
retransmit

a fixed number of times

Server crashes after receiving a request or reply message lost

(cannot be told apart!)
Client

resubmits request, server chooses:

- Re-execute procedure: service should be idempotent - Filter duplicates: server should hold on to results until acknowledged

Client crashes after sending a request

Orphan

detection: reincarnations, expirations

Reporting failures breaks transparency

Kangasharju: Distributed Systems 18

Fault tolerance measures

Retransmit Duplicate Re-execute/ retransmit request filtering no N/A N/A yes yes no yes re-execute retransmit reply

invocation semantics maybe at-leastonce at-mostonce

19

Kangasharju: Distributed Systems

CORBA
Interface Repository IDL Compiler Implementation Repository

Client

OBJ REF

in args operation() out args + return

Object (Servant)

IDL SKEL DII IDL STUBS ORB INTERFACE

DSI

Object Adapter GIOP/IIOP/ESIOPS

ORB CORE

CORBA shields applications from heterogeneous platform dependencies e.g., languages, operating systems, networking protocols, hardware
Kangasharju: Distributed Systems 20

XML RPC

Kangasharju: Distributed Systems

21

RPC: Different Systems

Kangasharju: Distributed Systems

22

Communication: Message Passing

Node Net Process A X=f(..); send X to B ... X: 10 OS kernel kernel Process B receive X from A Y=f(X); ... X: 5 Data Communication OS 10 procedure send buffer Network A<=>B procedure receive
xxxxx xxxxx xxxxx xxxxx

Kangasharju: Distributed Systems

23

Binding (1)
Structure of communication network
one-to-one

(two partners, one shared channel) (client-server) many-to-many (client-service; group

many-to-one one-to-many,

communication)

Types of message passing

send, on

multicast, broadcast

any channel structure

Kangasharju: Distributed Systems

24

Binding (2)
Time of binding
static

naming (at programming time) naming (at execution time)

dynamic

- explicit binding of channels - implicit binding through name service

Kangasharju: Distributed Systems

25

Persistence and Synchronicity in Communication

General organization of a communication system in which hosts are connected through a network

Kangasharju: Distributed Systems

26

Persistence and Synchronicity in Communication

Persistent communication

a submitted message is stored in the system until delivered to the receiver (the receiver may start later, the sender may stop earlier) a message is stored only as long as the sending and receiving applications are executing (the sender and the receiver must be executing in parallel)

Transient communication

Kangasharju: Distributed Systems

27

Persistence and Synchronicity in Communication

Persistent communication of letters back in the days of the Pony Express.

Kangasharju: Distributed Systems

28

Persistence and Synchronicity in Communication

Asynchronous communication

the sender continues immediately after submission the sender is blocked until - the message is stored at the receiving host (receiptbased synchrony) - the message is delivered to the receiver (delivery based) - the response has arrived (response based)

Synchronous communication

Kangasharju: Distributed Systems

29

Persistence and Synchronicity in Communication

a) b)

Persistent asynchronous communication Persistent synchronous communication

30

Kangasharju: Distributed Systems

Persistence and Synchronicity in Communication

c) d)

Transient asynchronous communication Receipt-based transient synchronous communication

Kangasharju: Distributed Systems

31

Persistence and Synchronicity in Communication

e) f)

Delivery-based transient synchronous communication at message delivery Response-based transient synchronous communication

Kangasharju: Distributed Systems

32

The Message-Passing Interface (MPI)

Traditional communication: sockets Platform of concern: high-performance multicomputers Issue: easy-to-use communication for applications Sockets? No: wrong level, non-suitable protocols a new message passing standard: MPI
designed no no

for parallel applications, transient communication

communication servers failures (worth to be recovered from)

Kangasharju: Distributed Systems

33

The Message-Passing Interface (MPI)

Primitive MPI_bsend MPI_send MPI_ssend MPI_sendrecv MPI_isend MPI_issend MPI_recv MPI_irecv Meaning Append outgoing message to a local send buffer Send a message and wait until copied to local or remote buffer Send a message and wait until receipt starts Send a message and wait for reply Pass reference to outgoing message, and continue Pass reference to outgoing message, and wait until receipt starts Receive a message; block if there are none Check if there is an incoming message, but do not block

Some of the most intuitive message-passing primitives of MPI.

Kangasharju: Distributed Systems 34

Message-Queuing Model (1)

2-26

Four combinations for loosely-coupled communications using queues.

Kangasharju: Distributed Systems

35

Message-Queuing Model

Primitive Put Get Poll Notify

Meaning Append a message to a specified queue Block until the specified queue is nonempty, and remove the first message Check a specified queue for messages, and remove the first. Never block. Install a handler to be called when a message is put into the specified queue.

Basic interface to a queue in a message-queuing system.

Kangasharju: Distributed Systems

36

General Architecture of a Message-Queuing System

The relationship between queue-level addressing and network-level addressing.

Kangasharju: Distributed Systems

37

General Architecture of a Message-Queuing System

2-29. The general organization of a message-queuing system with routers.

Kangasharju: Distributed Systems

38

Message oriented middleware

asynchronous messages

appl. A msg queue server

appl. B Q1 msg transfer system

SSL tms

reliable, fault-tolerant no loss, duplication, permutation, cluttering

persistent subscriptions models supported message queue request-response multicast publish-subscribe

msg queue server Q2

msg transfer system

appl. C
Kangasharju: Distributed Systems 39

MOM = message oriented middleware

Basic model: pipe between client and server
asynchronous

messaging natural, synchronous

communication cumbersome
message violates

queues support reliability of message transport

access transparency, no support for data

heterogeneity unless in programming language mapping, no support for transactions

suitable

for event notifications, publish/subscribe-based message queues support fault tolerance

architectures
persistent

Kangasharju: Distributed Systems

40

MOM Topics
Topics for variation and development
persistent/transient FIFO/priority translations abstractions

msgs

queues on msg ordering automatic load balancing

of msgs

multithreading, msg

routing (source, cost, changes in topology etc) transfer of msgs (at least between msg servers)

secure

Kangasharju: Distributed Systems

41

Message Brokers

The general organization of a message broker in a message-queuing system.

Kangasharju: Distributed Systems

42

CORBA Events & Notifications

Event namespace (names and attributes) Typed events (header+body; fixed + other) Consumer event filtering, event batching, event priority, event expiration, logging,
internationalization, flow control mechanism

consumer1 ... consumerN

event channel typed events filter n

QoS properties supplier1 ... supplierN constraints

43

Kangasharju: Distributed Systems

Publish-subscribe
shared mailbox, everyone can send to it subscribers can select what filter to use guaranteed delivery of all relevant messages to all
subscribers models: header-based, topic-based problems
scalability: ordering

comparing filters and messages

of messages

Kangasharju: Distributed Systems

44

Stream communication

Setting up a stream between two processes across a network.

Kangasharju: Distributed Systems 45

Specifying QoS (1)

Characteristics of the Input

Service Required

maximum data unit size (bytes) Token bucket rate (bytes/sec) Toke bucket size (bytes) Maximum transmission rate (bytes/sec)

Loss sensitivity (bytes) Loss interval (sec) Burst loss sensitivity (data units) Minimum delay noticed (sec) Maximum delay variation (sec) Quality of guarantee

A flow specification.

Kangasharju: Distributed Systems

46

Specifying QoS (2)

The principle of a token bucket algorithm.

Kangasharju: Distributed Systems

47

Setting Up a Stream

The basic organization of RSVP for resource reservation in a distributed

system.

Kangasharju: Distributed Systems

48

Synchronization Mechanisms (1)

The principle of explicit synchronization on the level data units.

Kangasharju: Distributed Systems 49

Synchronization Mechanisms (2)

2-41

The principle of synchronization as supported by high-level interfaces.

Kangasharju: Distributed Systems 50

Other forms of communication

Multicast (application level)
overlay

network where relays not members of group (tree,

mesh)

Gossip-based data dissemination

infect

other nodes with useful data by an epidemic algorithm exchange information with a random node infected, susceptible, data removed

periodically states:

Kangasharju: Distributed Systems

51

Chapter Summary
Overview of different interprocess communication
techniques and solutions Remote invocations (RPC etc.)

Message passing Streams Publish/subscribe Multicast (more on this later)

Kangasharju: Distributed Systems

52