UNI

TIV-PARALLELCOMPUTERSTRUCTURES

Par
all
elpr
ocessi
ngcanbedescr
ibedasacl
assoft
echni
queswhi
chenabl
est
he
sy
stem t
o achi
eve si
mul
taneous dat
a-pr
ocessi
ng t
asks t
o i
ncr
ease t
he
comput
ati
onalspeedofacomput
ersy
stem.Apar
all
elpr
ocessi
ngsy
stem cancar
ry
outsi
mul
taneousdat
a-pr
ocessi
ngt
oachi
evef
ast
erexecut
iont
ime.Thepur
poseof
par
all
elpr
ocessi
ngi
stospeedupacomput
erpr
ocessi
ngcapabi
l
ity
,andi
ncr
ease
t
hroughput
,ie,t
heamountofpr
ocessi
ngt
hatcanbeaccompl
i
sheddur
ingagi
ven
i
nter
val
oft
ime.Her
eweconsi
derpar
all
elpr
ocessi
ngundert
hef
oll
owi
ngt
opi
cs.

1.Pi
pel
i
nePr
ocessi
ng
2.Vect
orPr
ocessi
ng
3.Ar
rayPr
ocessor
s

Pi
pel
inePr
ocessi
ng:-Thet
erm Pi
pel
i
ningr
efer
stoat
echni
queofdecomposi
nga
sequent
ial
processi
ntosub-
oper
ati
ons,
wit
heachsub-
oper
ati
onbei
ngexecut
edi
na
dedi
cat
edsegmentt
hatoper
atesconcur
rent
lywi
thal
lot
hersegment
s.

Vect
orPr
ocessi
ng:-Av
ect
orpr
ocessori
sdesi
gnedf
orv
ect
orcomput
ati
ons.Av
ect
or
i
sanar
rayofoper
andsoft
hesamet
ype.Vect
orpr
ocessi
ngper
for
mst
hear
it
hmet
ic
oper
ati
onont
hel
argear
rayofi
nteger
sorf
loat
ing-
poi
ntnumber
.

Ar
rayPr
ocessor
s:-Ar
rayPr
ocessor
sper
for
mscomput
ati
onsonl
argear
ray
sofdat
a

Mul
ti
processi
ngsy
stem

A comput
er’
s capabi
l
ityt
o pr
ocess mor
ethan one t
ask si
mul
taneousl
yis
edmul
cal
l ti
processi
ng.A mul
ti
processi
ngoper
ati
ngsy
stem i
scapabl
eofr
unni
ng
manypr
ogr
amssi
mul
taneousl
y,andmostmoder
nnet
wor
koper
ati
ngsy
stems(
NOSs)
suppor
tmul
ti
processi
ng.

Adv
ant
agesofMul
ti
processorSy
stems
1.Morerel
iabl
eSy st
ems:
-Inamul
ti
processorsy
stem,ev
eni
fonepr
ocessorf
ail
s,t
he
sy
stem wil
lnothal
t.

2.Enhanced Throughput
:-The thr
oughputofthe sy
stem i
ncr
eases i
.e.numberof
processesget
ti
ngexecutedperuni
tofti
meincr
ease.
3.Mor
e Economi
c Sy
stems:
-Mul
ti
processorsy
stems ar
e cheapert
han si
ngl
e
pr
ocessorsy
stemsi
nthel
ongr
unbecauset
heyshar
ethedat
ast
orage,per
ipher
al
dev
ices,
powersuppl
i
eset
c.

Ar
chi
tect
uralCl
assi
fi
cat
ionschemes

Ther
ear
eav
ari
etyofway
sthatpar
all
elpr
ocessi
ngcanbecl
assi
fi
ed.I
tcanbe
consi
der
edf
rom t
hei
nter
nalor
gani
zat
ionoft
hepr
ocessor
,fr
om t
hei
nter
connect
ion
bet
weenpr
ocessor
s,orf
rom f
rom t
hef
low ofi
nfor
mat
iont
hrought
hesy
stem.M.
J.
Fl
ynnpr
oposedacl
assi
fi
cat
ionf
ort
heor
gani
zat
ionofacomput
ersy
stem byt
he
numberofi
nst
ruct
ionsanddat
ait
emst
hatar
emani
pul
atedsi
mul
taneousl
y.The
sequenceofi
nst
ruct
ionsr
eadf
rom memor
yconst
it
utesani
nst
ruct
ionst
ream.The
oper
ati
onsper
for
medont
hedat
aint
hepr
ocessorconst
it
uteadat
ast
ream.Thet
erm
st
ream r
efer
stot
hef
lowofi
nst
ruct
ionsordat
a.Par
all
elpr
ocessi
ngmayoccuri
nthe
i
nst
ruct
ionst
ream,
int
hedat
ast
ream,
orbot
h.Bot
hinst
ruct
ionsanddat
aar
efet
ched
f
rom memor
ymodul
es.I
nst
ruct
ionsar
edecodedbyt
hecont
roluni
twhi
chsends
decodedi
nst
ruct
ionst
ream t
othet
othepr
ocessoruni
tfort
heexecut
ion.Dat
ast
ream
f
lowbet
weenpr
ocessorandmemor
ybi
dir
ect
ional
.

Based on t
he mul
ti
ple dat
a and i
nst
ruct
ion st
reams Fl
ynn’
s cl
assi
fyt
he di
git
al
comput
ersi
nfourcat
egor
ies-

1.Si
ngl
einst
ruct
ionst
ream-
singl
edat
ast
ream (
SISD)
2.Si
ngl
einst
ruct
ionst
ream-
mul
ti
pledat
ast
ream (
SIMD)
3.Mul
ti
plei
nst
ruct
ionst
ream-
singl
edat
ast
ream (
MISD)
4.Mul
ti
plei
nst
ruct
ionst
ream-
mul
ti
pledat
ast
ream (
MIMD)
1.SISD:
Ithasonei
nst
ruct
ionst
ream onedat
ast
ream.I
tdoesonet
hingatat
ime.I
thas
capabi
l
ityofmani
pul
ati
ngonedat
ast
ream atat
imebyexecut
ingasi
ngl
einst
ruct
ion
st
ream.Mostser
ialcomput
ersar
ebasedonSI
SD.I
nst
ruct
ionsmaygetov
erl
apped
dur
ingt
hei
rexecut
ion.MostSI
SDcomput
ersar
epi
pel
i
ned.Par
all
elpr
ocessi
ngcanbe
achi
evedbymeansofpi
pel
i
nepr
ocessi
ngormul
ti
plef
unct
ionaluni
ts.SI
SDmayhav
e
mor
ethanonef
unct
ional
uni
ts,
whi
char
esuper
visedbycont
rol
uni
t.
Forexampl
e-I
BM 370comput
ers.
 CU–Cont rol Uni
t
PU/
PE–Pr ocessorUnit/Pr
ocessorEl
ement
MU-Memor yUnit
MM-Memor yModul e
SM-SharedMemor y
I
S–I nst
ructionstream
DS–Dat astream

2.SIMD:
SIMD comput
ers hav
e si
ngl
e i
nst
ruct
ion
st
ream t
oexecut
eonmul
ti
pledat
ast
ream.I
thasasi
ngl
econt
roluni
ttogener
ateone
i
nst
ruct
ionst
ream atat
ime.Asi
ngl
econt
roluni
thav
emul
ti
pleALUs(
Ari
thmet
icand
l
ogi
cuni
ts)t
owor
konmul
ti
pledat
ast
reamssi
mul
taneousl
y.I
thascapabi
l
ityt
o
execut
easi
ngl
einst
ruct
ionst
ream onmul
ti
pledat
ast
reams.I
tsal
soknownasv
ect
or
orar
raypr
ocessor
smachi
ne.I
nSI
MDmul
ti
plepr
ocessi
nguni
tsar
esuper
visedbya
si
ngl
econt
roluni
t.Al
lpr
ocessoruni
tsr
ecei
vesamei
nst
ruct
ionf
rom t
hecont
roluni
t
butoper
ateondi
ff
erenti
temsondat
a.Theshar
edmemor
y(SM)uni
tsmustcont
ain
mul
ti
plemodul
es,
soi
tcancommuni
cat
ewi
thal
lpr
ocessor
ssi
mul
taneousl
y.
Forexampl
e-I
LLI
AC-
IV

3.MISD:
MISDcomput
ershav
emul
ti
plei
nst
ruct
ionst
ream t
oexecut
eonsi
ngl
edat
ast
ream.
Thi
sty
peofsy
stem i
snott
obui
l
dpr
act
ical
l
y,i
t’
sat
heor
eti
calappr
oach.I
thas
mul
ti
plei
nst
ruct
ionst
ream,whi
choper
ateonsamedat
ast
ream.Theout
putofone
pr
ocessorbecomet
hei
nputofnextpr
ocessor
.
4.MIMD-
MIMD comput
ershav
emul
ti
plei
nst
ruct
ion st
ream t
o execut
eon mul
ti
pledat
a
st
ream.I
thascapabi
l
ityofper
for
mingsev
eralpr
ogr
amssi
mul
taneousl
y.I
tissi
mil
ar
t
omul
ti
processor
,inwhi
chmul
ti
pleCPUsar
eoper
ati
ngi
ndependent
lyt
obeapar
tof
l
argesy
stem.Bot
hmul
ti
processorandmul
ticomput
ercomesunderMI
MD.When
mul
ti
pleSI
SDwor
kst
oget
hert
hani
tscal
l
edMSI
SD,whi
chcomesundercat
egor
yof
MI
MD.I
fnumberofi
nst
ruct
ionsar
ehi
ght
hani
t’
sknownast
ight
lycoupl
edel
seknown
as l
oosel
y coupl
ed.