You are on page 1of 15

menoy Ceitral as woksdtina,

pocesig;
dytbte!
L,LANs
Shecs Newok
Tndesonneeten Conputaon
pasui
oMesaye-
y Tree
Fat
Guey/ hph
Mesh
Conneced high
slesns fer
Hypercabes & onoem
DEC,
arehhe archie
MIMD LANsavilbilt
clstes. addunbboce
doe PE tultionptes Fntel
multieompters
NCUBE
MS bFost
Distubted
Cscalab
le) pCs
mulbe 0M
contoly
hoad
momymeny.
procein muliple
oton. Compates. Compukyhoiny
ove Hitachi,
yjitou,
Czay, be
mutisye
bullel muti, HiMD obpea
ithdbtbted
sheoms aplie calatble)
Cnet
Centrol
memany Tera
NCR,Sun
DEC,
EnQne,
MIMD ohoit
Crass ing SGI, muApoces memoues
Compuliny,
Computiny MIMD toxonomy to
Simple, Sequend, yet
tuatbn
asheom tomputaton
wemoy menay wit
Taxanovay a ditibuted
Unseoloble ae
mltipocessOs
use
anle
addes
poce shared-mtmery
provideo ble
distibuded
þeyfotmone KSR multhposel
dao Mulirocass poesaos boce hared
sealo
(lbe Dutibted
memay eocheing bindiny
puopAom
stahe has
Bells +Oymomicing stondod
afbinding GelC1992) wse
dfendhijh CM
Cedo,
B8N, addhs a
un binding, ASHAW'ont,
bindiy,
Hlcomdes
Hultieombu
Gondon taddhenes
o consideng bwted
MMO usedota (292) Shaod SC1
Gondon wsimyleduti
oh sadie
IEEE besote
N He a
- dt
ACA
Flymn's Cloaifieshon 1922
OSISD SIMD 6) MIsD mTmo
Siyle Jay uliph doo
ACA ’ stdy biplemuntn comex ond sobhiscth'eteo tu
ems,

SIS) ( SIMO
Siyle TodracnnSiyleDAtasteomm Trodsuehin eulhhla doto
Siyle
CFrstueettn Poot
(ab)

2+3

|2

NISD
MMD

Kaxt)a

MISD
Uemg
DS

MIMD Shuet93

dy'ntdes data,
3) emy Tdsenol

both Ooynche Greuti


s)-Task llbm
Mlivet & SMp Combutey
Veeto bhoeene
-siry le
deverel biecas
doto c'th iye artwcin
opeates on aaey doa foin
Csch inastucton hndled bepunclly -pnulie dems has hove
Jse mme instet'on Con be com,bed

Selon Proceser
Seler Funebonal
fpelines
Sealar stesion Vectoc Cescess a
Stelor 0veston e ter
lontal Unit ingud Coituol lUnii

vecta Vecdo Fn
Spalor Main Memony Vecdo

Mos Het
Reploce sitn Vedor Steamiy Unit
|Stoe for memay anch'te

24 Vocto procmen gtachud do deoler ocsn as


oll inhut fiyd docode by ela Contial t

E. Conea C3%o0 UA*900à Sse


lesees
veder CU
SIMD
Conhal Unit
PEo PE PEL

MemO rem

Tn terconheton Netwt
SiMp Canþutes
Simo Corp. hovnp SeThst eve maliple doto steeons

M=CN, c, I, M, R) CneIM
N- No. PE in re machine llocey clom)
2)
Frelud

dco exe cuded by Cu


) I- Ses Thd. boa dcasted b CU to all fEs for þodll ereaton
( indudy ahmehe , lqie,'ment'y ete)

) R- Set dta auty funtons (on nten fE Commnl'e)


Masfer M-1, Cm-2

PRAM & VLSIMeolel


l compters absubed o Jhyics mol
NUna, UmA
hese e ubed

PRAM ( arallel Rondom Access Hackie)


Fantne ond wylie (1493)
dor mad eliy ideoized lel compteas oith
zer0 anehonisoton memony acces oveahend
llsed
hicitcal model in lel compaiy
PRAM mod impidiesnoos hy asm on delsed lel
Baie lanabti
shorad menoy nchdehe
oc ens do Rsbel
Contaal Fantne
uyllie jndonal Shered
(948)
Global Heriary
PRAM odels
PRAy
-Cnocear 'oed na pocess (ons, eoch with its own lontel Unit
Eroceses oheeste con cuny Cosm timn e irdtoa don

memey henttine into cellu,


Wihshoned msm ony, Foun Aenoy-Uhdete apobns ane
posibte
Excluse Read
(ER)
allows at rmest one pocaMOr to heod
hom ony emmey Jocason 1h eoch cyole.
Exlive te- allowb ad mast ane proeson so Jte
(Ew) jnto a men Jocot'on ad o m e
Concuwsent raad- ollos t e dmltipsceou to ead Ju soma nfetmtn
(CR) tom beme mlm oy call in e some cyele
(cw)
PRAM (dopendiy ypon boy the mtm ay Jaeds & ites
) EREW 2e honddle

CREw
bt ony one con wnite t odme Cmtl erlion)
ERCw Alows enclusive cheod Cances t
to sem memy loe
9CRes some hme
Confly Writes
() CommonAl
ane esalued by
imeltoneus wye stare Je som Value

3) Miimum- Vole ten by koocUsa Ly, Do nimum ndo el oman

& no rocesoo lsed

ho Cost

# VLSI Comlenig Hsslal lAT Madl)


The VLsI (Very leye Seda htepradbn) comploi t model is
eheseurces chaâuind to implemat o poticle o & Corputoton
On VLSI hip
* T foeuses en hardwne
hodwne apeca
aspecs Tc despm, senaidauiya-g
aspet Consideuy elemadh ueh
her Conne
tel comptes

B3 lower beeudo on VLSI eûcud in terpnetedy utsmen (1934)


en
meey, Tlo &camunieton

Lo9 A be the ch
Ilo bond : AT Communieain-mted bound JAT
Time

AV
( Cmunless litel bound
OMomayrm ed band on
chip aee

chip histoy apeJontes


dh volume AT
CM2(Comeston Mochine-) in, aMPtompater ftayubr datlai
The
farelel seaportompute computer )

speed g obove do Gopa.


okieviy ted paconoiy

Sepuence Seslu meno} Gus


Gbl Res Bas Trstuo Brokes Bu

fCombine

Theundo
ehocsaw.
E
Homenies M m

RoterINEWS |Sconniy

Conttdllen
Tlo
Conlea
Eumsballer

a font end royh Seancet


hoyh rtuction Boedeegd Bus do oll doto poce.

nenoinstins eceros dhoyh beyuwnc. proess lon he


bineanse
Coeenoeo eschuye hemsel in lee dhoy he rsr,
NES Sconni
d en michon m.
Thuse ae Canneded o Flo indefes Amos sbye Jubdyse

allos fnd end oo dood ono wte onu 395* volg


Dopleachbna!
-onlliaye Jvision ocassiy &ine podo n
Neuel natwa
Contod ee peay

d dhemtes

# Vecto tns tuton ytes:


rulito desda.
Veetan - e o notuetes

Veton -cele matutm.

Vedo -memer
Vetu lood
V,
Veet steee
veetn- eduton hat
de Ve’j

S) Gether te tt

Seoter
(unh)
Moshiy
#leco-postin Teahn'ues

atoncemes y memye- tsiy eemmly


On lane, stahe dota stude
elmd, dhun
hun One

Domah Decomposshon : she d,stn


naplenty
invored.

O Stade doto stutre: Fo y: mtis foctarizator for dalig o


- diesne peclen jote, ith o
loye fite
ynole det stuua ted d syle entity: For f, in a dlat
duodls Jru badie cn be didihs to
mog-body polem, atond Joes, th bo dies mey be mteaoiy oih each
nodis. Taut povid stay on
other ond movig in pace. The caluletin fon eoch body con
Fined domain oith dyomie ornbutatons th vovian fions ghdanoin
fae cf a Aiopom t model dyneie id vortces hoho
move around

Thres ebo o decompebe "domoin A'ren obþhee


> Dsdaibate obdomsin d de, to Vovs nodes
he
Rsrid She Compashon so shet eoch node anopsom apdots

in nade hoeps s.
po tesn qpum do ve'aus hode
Compile & un On Gre node on
un mlso copis deme proynon mang nodes ad ena
dato
E: LINACK nahiy bteiAta siy domun dacanfott
R=
dimonl mn
nmn

otone R indto two Jove nah moves U(mrt) sutxn)


Un
U2 U V2s Vn
Umt
Spahel V Vsa
O (aat Donoh Decoposton,
Di'de m h

Tndopnu foouzehon
eoh ocesa

co mmunicot

meda
Decmprt

do-thol decomt: used an symbale procnsig pcbkms. in AItMes


Funet'on Ceeomeston:
Imye enhoncm

blsu
Fnge enhines , de deten,
mod
edse deted

dh mads

Monye Uatn cthro4ch


dinoe & nl s dehnie
dinle ohblaston tests.
Cese one y dhe pno tesoes esonper noe hast 0bn node

oay ete
ben

beyon desk
N

Comhenleymonya betee oludon & lonbis thm to


Cnpek sel
wortn no o meny
Fro Ses heh,'teate
We

Chattei
Cooper

Voislee-644) Jor
ben

Go fefsse S-4 GPRA, 39-19 2 GPR loye umders


Rayue GpR Unyed coche fant ddo blit doto coehe & nstCoche
OI8loct 83-So MH2 So-lSo MHe
Rte
micRo a do d wiy cemeg

( RISc
Kodoed

Coen doteesehe
clos.
loy
em op! FPa

loo

Desodontyn
Trest oche
areas, incud.
Symbolic Processors Symbolic processing has been applied in many
engineering, text
ing theorem proving, pattern recognition, expert systems, knowledge and
Tetrieval, cognitive science, and machine inteligence. In thesé applications, data,
IO
knowledge representations, primitive operations. allgorithmic behavior, memory,
and communicatons, and special architectural features are different than in numerical
Comnnting Symbolicprocessors have also been called prolog processors, Lisp processors,
or symbolicmanipulators.

caption Characteristics of Symbolic Processing


Attributes Characteristics
Knowledge Lists, relational databases, scripts, semantic nets, frames,
Representations blackboards,objects, production systems.
Common Search, sort, pattern matching, filtering, contexts, partitions,
Operations transitive closures, unification, text retrieval, set operations,
reasoning.
Memory Large memory with intensive access pattern. Addressing is often
Requirements content-based. Locality of reference may not hold.
Communication Message traffic varies in size and destination; granularity and
Patterns format of message units change with applications.
Properties of Nondeterministic, possibly parallel and distributed computations.
Algorithms Data dependences may be global and irregular in pattern
granularity. and
Input/Output User-guided programs; intelligent person-machine interfaces;
requirements inputs can be graphical and audio as well as
to very large on-line databases. from keyboard; access
Architecture Parallel update of large knowledge bases,
Features dynamic memory allocation; dynamic load balancing;
hardware-supported
collection; stack processor architecture: garbage
ecture; symbolic processors.
&
deess
h'me Cost/b
inhe
coteeh
Heer ey
Merag
men d R s
Mn $s D hdd
3
e

mcg
nat be i
wora ms

cobies beme ino es due en ma


hy

tondeng Alen
temhanl tocdy nelon) oceey

You might also like