Parallelism

Purposes

Speed up
Example: Supermarket check-stands
Reliability
Example:Big-rig tire configuration
Logical organization
Example:UNIX^®

Biological Parallelism

(Neural Networks)

Humans are smarter than computers.

Yet human neurons switch at rate of about

1 per 10^-3 sec

while computer logic elements switch at rate of

1 per 10^-9 sec,

10⁶x faster

Human brain has about

10¹⁰ neurons

and 10¹³ connections between them.

Computer has a few million logic gates.

Humans are smarter because their neurons work in a more highly parallel fashion.

Parallelism can be exploited at many levels.

Logic Level
Instruction Execution Level
Process Level
Thread Level
Cluster/Network Level

Logic Level

Instruction Execution Level

Process Level

A process is a program in execution.

Many processes can be run concurrently:

muliplexed among 1 processor

or

N processors

Processes by default don't share memory:

Each has its own address space

Communication is via pipes, sockets, ...

Processes on Unix Command Line

p1 & p2 & p3

p1 | p2 | p3

Processes in C code

fork( )

(returns id of child, or 0 if child)

Thread Level

A thread is similar to a process, but finer grain, lighter weight.

Threads are within a process, do share memory and other stuff.

Cluster/Network Level

Processes on different machines, possibly in different locations, communicate.

Languages supporting Threads

Java-built-in

rex

C/C++ depending on operating system

rex threads

Solaris threads (muddcs)

Real vs. Virtual Parallelism

Real: Separate physical processors

Virtual:

Processes are multiplexedamong fewer processors

(e.g. 1 processor)

Java threads(see Ch 8, pp 210,211)

2 models:

(1) is simpler, easier to understand

(2) is used in all most Applets

Once you see (1), it is easier to see (2).

extends Thread model

A class is defined which inherits from Thread.
When a new object of the class is created, a thread is automatically created.
Calling the run method of the class runs a thread in parallel with the creating thread.

See BouncingBalls (source BouncingBalls.java) for example.

Bouncing Balls

Ball extends Thread
Each ball executes its own run method (same for all balls).

Each has its own state.

implements Runnable model

Class Ball implements Runnable
Must define 'run' method (other methods possible too)
Object of class Ball is created.
A thread is created which "runs" the object in parallel with other threads (by calling its run method)

See source AltBouncingBalls.java for example.

Synchronization

Threads can get into each other's way:

one changes variable while other is reading

Difficult debugging

Some tools:

Network-level Parallelism

The network is the computer.

Sun Microsystems

Example: Client/Server

Communication over network takes place via data abstractioncalled "socket".

Each networked machine provides set of logical, numbered, ports.

One machine (client) can establish communication with another (server) by knowing:

(a)server's internet address

(b) server's port

Server is running a logical process "listening" at a port.
An incoming message designates a client wishes to connect.
When server acknowledges client, a new process (or thread) can be created to carry on a dialog.

Client/Server Code

(from "Java in a Nutshell" book)

Port # is agreed upon in advance ( 6789 )

Server:

Client:

Example

Server is running on turing at port 6789.

Connect from clients on turing, muddcs, or orion

cd ~cs/cs60/java

java Client turing

(host name)

connected to 173.134....

> foobar

raboof

> ....