Elenco Electronics MO-1251 Manual de usuario descargar pdf (Pagina 80)

86

Parallel

architectures

In

conventional

processing,

the

system is

fed a problem

se-

rially,

one

chunk

at a time: do A,

then

B, and then

C. An

advanced

architecture,

called

a concurrent

system,

might attack

the prob-

lem

by accepting

instructions

A.

B.

and

C simultaneously,

and

then

routing

each instruction

to

a separate

processor

that

acts

upon it.

Planning

and publishing

a

magazine

provides

a good analo-

gy. A magazine

is basically

an an-

thology

of related

articles

placed

under

one

cover. To produce

a

magazine,

the publisher

draws

from many

sources, including

authors, news

services,

public re-

lation

firms, and

advertisers.

Be-

cause the information

is

gener-

ated by each

contributor

inde-

pendently,

much

of it can be

done

simultaneously

without the

work

of one

depending

on the result

of

another.

In effect, work

proceeds

concurrently

until

the results

are

collected

and assembled

into

print.

Similarly,

a concurrent

system

works

on more

than one

aspect of

Companies

Mentioned

Intel

Corporation,

3065

Bowers,

Santa Clara,

CA 95051.

(408) 987 -8080.

Number

Nine Computer

Cor-

poration.

725

Concord

Ave.,

Cambridge,

MA 02138.

(617)

492 -0999.

INI.

Computer

Systems,

6915

Hightech

Dr.,

Midvale,

UT

84047. (801)

561

-1100.

MicroWay,

Inc.,

P.O. Box 79,

Kingston,

MA

02364. (617)

746 -7341.

Computer

System

Archi-

tects, 950

N. University

Ave..

Provo,

UT 84604. (

801 )

374 -2300

Definicon,

1100 Business

Center Circle,

Newbury Park.

CA 91320.

(805) 499

-3559.

Intel

Scientific

Computers,

15201

N.W. Greenbrier

Pkwy.,

Beaverton,

OR

97006.

(503)

629 -7631.

Autodesk,

Inc.,

2320 Mari -

nship

Way,

Sausalito.

CA

94965.

(408)

332 -2344.

CV)

a

b

c

Mk,VJA

1b1417

h

Fig.

4. SEVERAL

witr

KKLUIV/VrC, l

t KA/VJYU

I EKS:

(a)

linear. (b)

_fast

linear. (c)

2D

array. (d)

cube. (e). cross-

connected cube.

(J) 3D

cube. (g)

cross-

connected

3D

cube. (h)

hypercube.

and (i) 64 -node

torus.

a problem

at

once.

First the prob-

lem

is broken

down into

as many

parallel

elements

as possible:

then

each processor

in

the sys-

tem is

given

an assignment.

When

all processors

have re-

ported

in, the results

are as-

sembled

and analyzed.

Based on

those results,

the

software then

initiates

another

round of

pro-

cessing.

That

winnowing process

continues

until a

solution to

the

overall

problem

is found:

the

number

of steps

required

de-

pends

on the number of parallel

processors

the software

supports

and

the magnitude

or complexity

of the problem.

There

are many ways

to assem-

ble a concurrent

system.

some of

which are

shown in Fig.

4.

Parallel programming

Because

conventional pro-

gramming

languages are not

de-

signed

to deal with parallel

processors,

alternative parallel -

speaking languages

have been

developed.

Some are

esoteric lan-

guages

written

for

a specific

ma-

chine

in

a specific

configuration,

but

many

are adaptations

of

more

familiar

languages.

11.ansputer

boards

are gener-

ally programmed

in

modified

ver-

sions

of

standard

languages,

such

as Pascal,

C. and Fortran.

One of

the more

popular

trans -

puter

languages,

Occam.

is akin

to Pascal.

Except

for

the paral-

lelism

commands,

working

with

that type

of language

is similar

to

working

with

a

standard

dialect.

For

each

parallel

system,

how-

ever,

a different

software

applica-

tion

that recognizes

the number

of transputer

chips

and

their

configuration

must

be written.

You place

that information

in

a

header

library

that

precedes

the

run

library

(the

program

itself).

After

compilation,

the program

will

only run

on that

system

or an

identically

configured

system.

Not

one

change

can

be made

to

the transputer

hardware

without

going

back

to the

original pro

-

(Continued

on page

94)

1 2 ... 75 76 77 78 79 80 81 82 83 84 85 ... 99 100

Comentarios a estos manuales

Sin comentarios