Parallel systemhhzgzhzbzhhzhzhuzhzhzhhzhzh

Parallel & Distributed
Systems
Communication

Importance of
Communication Network
 Performance of a distributed system
deeply depends upon underlying
interprocess communication done
through communication network
2

Networks
• Definitions
 A set of Machines (computers, phones, routers, switches,
gateways, etc.) connected by communications links in a
way that they could pass information between each
other
• Why Networks?
 Sharing resources
 Increasing proximity between people/machines

Requirements
• Performance
 Transmission delay is arbitrary but finite
 Transmission time

Latency + message size / data transfer rate
• Reliability
 Hardware medium does not support replication
 Error recovery

Message lost

Message duplication

Message out-of-order

Message corruption

History of Networking
• Initially circuit switching (PSTN)
• 1960’s: Packet Switching
– ARPAnet (1969)
• 1970’s: LAN
– Ethernet
• 1980’s: Workstations & PCs
– Proliferation of LANs & WANs
• 1990’s: WWW
– Explosion of Internet nodes
– Mobile networks

Network Basics
• Topologies
 star, bus, mesh, ring, etc.
• Switching
 circuit, packet, and virtual circuit switching
• Transmission technology
 point-to-point vs. broadcasting
• Types of networks
 LAN, MAN, and WAN
• Internetworking devices
 Router
 Gateway
 bridge

Protocols
• A protocol specifies
 rules and formats used for interactions between
two parties

specification of the sequences of messages

specification of the format of the data in the message

Layered Architecture
• Motivation
 reducing complexity by modularizing tasks vertically
 Each layer use and provide services

layer N provides a service to layer N+1

layer N extends the service of layer N-1
 independence of each layer allows various
implementations across layers

Services in Layered
Protocols
• Data Transport Service types
 Connection-oriented vs. connectionless
 Virtual circuit vs. datagram
 Depends on data and QoS requirements of layer above
• Packet assembly
 message at layer N+1 is fragmented into multiple PDUs at layer N
(MTU at layer N+1 > MTU at layer N)
 layer N at peer: responsible for assembly
• Addressing
 identification of the peer at layer N
 conversion into layer N-1 address

Protocol Design Issues
• Quality of service
 reliable vs. unreliable transmission
 messages vs. stream
 error-free vs. error-durable
• Error recovery
 data errors

causes: network error, malicious attacks, etc.

detection and correction: CRC
 transmission errors

causes: network failure, overflow, receiver busy, etc.

detection and correction: sequence number and retransmission
• Flow control
 sliding window

Layers
 Low-level layers
 Transport layer
 Application layer
 Middleware layer

Low Level Layers
• Physical layer: contains the specification and
implementation of bits, and their transmission
between sender and receiver
• Data link layer: prescribes the transmission of a
series of bits into a frame to allow for error and
flow control
• Network layer: describes how packets in a
network of computers are to be routed.

Transport Layer
• The transport layer provides the actual
communication facilities for most distributed
systems.
• Many application protocols are directly
implemented on top of transport protocols, doing a
lot of application-independent work.
• Standard Internet protocols:

TCP: connection-oriented, reliable, stream-oriented
communication
 UDP: unreliable (best-effort) datagram communication

Session and Presentation
 Session
 Dialog control
 Synchronization
 Presentation
 Concerned with meaning of bits

How data is presented so that both parties can
understand it
 Inter-conversion (ASCII, EBCDIC, Unicode etc)
 Compression/encoding/encryption
15

Application Protocols
 The protocols that do not fit into one
of the underlying layers
 All distributed systems are just
application (from OSI perspective)
16

Middleware
 An application that logically lives in the
application layer but contains general
purpose protocols that warrant their own
layers, independent of others

Authentication and authorization protocols
 Naming Protocols

Not closely tied to any application

Can be integrated as a general service

Application independent nature
17

Inter-Process
Communication
 Communication between two
processes
 Required for distributed application
operation
 Message Passing – the simplest form
 Sending process to send a single
message (datagram) to a receiving
process asking for services
18

Inter-Process
Communication
 Allows communication between
components (processes)
 Shields one process from failure of
another
 Provides modularity by a well defined
interface mechanism
 Hides distinction between local and
remote communications
19

Design Considerations in IPC
• Data representation
• Marshalling
• Calling semantics
• Addressing
• Reliable delivery
• Versatility
20

Data Representation
• Due to heterogeneity property of distributed
systems
• Things to consider
 Representation of floating point number in different
architectures
 byte ordering: may be big endian or little endian
 word boundary: ASCII take 1 byte per character, Unicode
take 2 bytes
• Policy
 conversion to the common data type agreed by both a
sender and a receiver while passing parameters in a
function call from sending to receiving process and same in
return value
21

Data Representation
Examples
22
Standard What it does
Sun XDR (External Data
Representation)
Used in Remote Procedure Calls
(RPC); converts data into a
standard binary format.
ISO ASN.1 (Abstract Syntax
Notation One)
Common in telecommunications
and networking; allows structured
data exchange.
Xerox Courier
Early protocol for data exchange;
used by Xerox’s networked
systems.

Marshalling
 Process of taking a collection of data items and assembling them
into a form suitable for transmission in a message
 Translation of structured data items and primitive values into an
External Data Representation (XDR)
 Unmarshalling is process of disassembling them on arrival to
produce an equivalent collection of data items at the destination
 Both are intended to be carried out by a middleware layer
without any involvement on part of the application programmer
 CORBA’s common data representation
 Java’s Object Serialization
 Marshalling into binary/ASCII form

Used by HTTP
23

Calling Semantics
• Remote operation invocation may need a different
behavior than a local one

sender may not need a result either at all or immediately

sender may want to do other operations in parallel (e.g.
multithreading)
• Semantics
 synchronous: same as local operation invocation
(blocked)

timeout is used to avoid indefinite wait
 delayed synchronous: the sender may get a reply later
 asynchronous: the sender does not need a reply
 RPC Exchange Protocol

R(Request)/RR(Request;Reply)/RRA(Request;Reply;ACK)
24

Addressing
 Identification of communication peers
 Location independent identifiers are
required

Globally unique identifier: UUID in Network
Computing Architecture

UUID is Universally Unique Identifier

A URI

Examples of Identifiers

Ports

Mailbox
25

Reliable Delivery
 Reliable Delivery
 End-to-end argument

error recovery in lower levels of protocols is only useful for
purpose of increasing efficiency

protocols for application-level end-to-end checking are always
required
 Possible causes for retransmission (client -> server)

server still working on it

server crashes

request gets lost
 IPC should provide some level of failure transparent but
also failure visibility (accurate report of failures)
26

Reliable Delivery
 Retransmission mechanisms
 reply can be considered an acknowledgement
 selective retransmission: sequence number
 explicit acknowledgement (server/client)
 retransmission timer
27

Versatility
 Types of IPC

remote operation

bulk data transfer

group communication

continuous media
 Flow control
 stop-and-wait

sliding window
 QoS
 real-time with no error tolerance

stream-oriented: steady and low delay but packet lost is ok
28

Group Communication
 Pair wise communication is not best in all
cases
 A process speaking to group of processes

May be replicas
 A multicast operation is more appropriate
 A single message from a process is propagated
to all the members of the group

Group membership is usually transparent to user
29

Group Communication
 Usage

Fault tolerance based on replicated services
 Finding discovery services in spontaneous
networking

Router solicitations and advertisements

Better performance through replicated data
 Propagation of event notification

A new news message
 Example
 IP multicast
30

Case Study: IPC in UNIX
 Implemented as system calls in a layer over TCP and
UDP
 Uses sockets
 Messages are queued at receiver socket until
receiver process makes an appropriate system call
 Socket creation
 Any process can do this by invoking socket system call
 Primitive Arguments

Type: datagram/stream

Protocol: TCP/UDP
 Socket call return a descriptor
31

Case Study: IPC in UNIX
 Datagram communication
 Socket pair is identified each time a
communication is made
 Stream Communication
 A connection is established between
peer pair of sockets prior to any
communication
32

Parallel systemhhzgzhzbzhhzhzhuzhzhzhhzhzh

More Related Content

Similar to Parallel systemhhzgzhzbzhhzhzhuzhzhzhhzhzh (20)

Recently uploaded (20)

Parallel systemhhzgzhzbzhhzhzhuzhzhzhhzhzh

Editor's Notes