' ..

I
l I n tr o d u c ti o n

l\7 !'~ ; ;h, m R.., f,~rit 'n 1t> \ \rr ttp mt•r \
\~~ft•m t ' • u11pot1£"nl.i1 Mo
,\ ! ,'13:.~ t~ J'lc t.,;;,,;;1 ~
,\ \ ~tt<'-»M ,x\ rsNS ll\l ot lon - Re /a lio n to pa
sh 11 rrd ra / if;tl sys tem., _
~, --.'"tt1'1:"';~ i ,'u tr, vi ,\'\ 'nd mtJ mO tJ sy.ftom.t
•""el.
1
ih m,)HJ , .._,1·.~u .t <1,~y ncl,r0 Pr im iti ve s for dis tri bu
11ow , exec 11tl mu De slR ted
'lf .li.$,triJn,te>J t"r>-,,y>1m,rirttt,.~ n l.,s uc s an d challenges.
A
~..,.\ iY id s <f iYt•.,.rmmrirotiott : A dis tdb u tt:d pr og ra m
rw t\i't)r•k.~ - Glt1bul .t trJ" A m od el of dis tri bu ted ex
ec utions
-.,\1.,n-ls \~f r,1\)<'f;'...\'S t\)fflJmmi<" ! - Cu ts - Pm t an
d fu ture co nes of an eve nt
-1licms. L<>Rknl 11mt
~ ~".t1.'tT :im.· - J'(t' t.) r tint~· : A fra m ew or k fn r a sy ste
• l'/ t,wic-al do rk sy 11d 1ro m of log ica l clo ck s
n i2o
tio 11: NT P.
C oa ten ts
r. f Definition of Di str ibu te d Sy ste m
, .2 Re lst ion to Co m pu te r Sy
stem Components
1.v~ Mo
· tfv 8t ion
.· . . . .. .. . . ·. .. . . . .. . De
c.-18, · · · · · · · · · · · · ·
1. .f Re lation to Pa ra lle l Sy · · Ma rks 8
ste ms
;_5 M es sage -P as sin g Sy
ste ms ve rs us Shared Me mo ry
Systems
1. 0-Pr im itiv es fo r Di str ibu te
d Co mm un ica tio n
1.1 Sy nc hr on ou s ve rs us
As yn ch ro no us Execution
s
1.8 De sign iss ue s an
d Ch all en ge s
1.9 A Model of Di str ibu
te d Co mp ut at ion s : A Di
stributed Program
1. 1o A M od el of Di str
ibu te d Executio_ns
1. 11 Models of Co m m un
ica tio n Ne tw or ks
1. 12 Gtob al St at e
1.13 Pa st an d Fu tu re
Co ne s of an Ev en t
1_1 4 M od 6/s of Pr oc es s Co mm un ica tio
ns
1. 15 ClocJc Ev en ts an d Pr
oc es s State
1.16 Logical Ti m e
1. 17 Ph ys ica l Cl oc k Sy Ma -1 5 · · · · · · · · · . . . . . . ua rk s 8
nc hr on iza tio n : NT P . .
..... Y , 'VI
Two M ar ks Que-stions wi
th An sw er s l
(1 - 1)
 ,,,, ..,.,
,,

"f
'. i
,
hn ll\linframet• Low price /performance rAtio i1 the
meu, computtng power,
l08t
'
t ..._.. 1 A collection of mim>pl'O('•NOl'I ofter " hettr.r prtc,1/pnt,irn
effc, 11 .,,1, i l,~.
w,y
~

\ lpNd I A dlatrtbuttd ,yattm may havt mort1 total computinK power lliar,
llllllmllne. f

l Diatnbulld ayattma bt oxttnded through the addition of r.ompn,,t'n~

CII\
\Nnby providing bttter ecalabllity compartd to centrahzrd Aystt•ma. '
4. ...... dlltdbatlon I Some application, are lnhertntly distributed e
81
9'lpllffilrat chain.
S. W.bUlty I II ont machlnt cruhle, tht l)'lttm u a whole can 1ilill li11rvivP, 11
fV11 higher availability and Improved reliability.
I. ~ powth I Computifta powtr can bt added in small incrcmcntR.

- ...... of Dlltrlbuted lyattm

• IHhlD lhlrlnc ii main motivation ol the diatrtbuted 1y1ttm. The term ''rc11oum'
ta i l'llh1r lbatnct OM, but It belt charlCtldul tht range of things that can
1111fdy N lhutd II\ I Mtworktd computer l)'ltlm,

• • 11'1 be tM eoftwut l'IIOUIClt or hardwue ruourcn. Printers, disb,

~~ and data IN lht example of IOftwart and hardware l'flOurces. Sharing of
• • • . . . . from Nrdwart compon,nta 1uch u diw and printers to
• • • ~ d6ild .tnlltltl ll&Ch u ftltt, databaNa and data objectl of all kinds.
..... 1M 111111n of vidto frllNI 11\d audio connection that a mobile
----o1.,....typt..,.-111
Ill• A NIOUl'et mlNpl' II a IOftware module that manages a

■~
1111•11i1 u followt:
a COlllllllrlq o1np1tcatect data
.._.-, 5. Concurrent transactions
 1• 5 Introduction

'-'ghtpmenta
... ,... are shared to redu.cc cost. I) atn 8h:m:d .m dntabasc or web p·igcs are
ru •Q:'.Ve1 re"ources which arc, more signlf1cant
. to users without regard ' for the
tefYer or servers that provido tlw.sc.
, Types of resources :
.
1. Hardware n·~ourcl' •· 1fa• tll lti s,.. . , prmte,, i:-;uncrn, sc:ann,.r
1 Data : File, d:\t,lbns~, weh pngl'.
3. Service : Search l'ngin-.?
• Patterns of rcsourrc sh,mng vary widely in their scope and in how closely users
work together :
1. Search Engine : Users 1\t'l'd no contact betwc~n users.
2. Computer Supported Co-operative Working (CSCW) : Users cooperate directly
share resources. Mechanisms to coordinate user's action are determined by the
pattern of sharing and the geographic distribution.
• For effective sharing, each resource must be managed by a program that offers a
communication interface enabling the resource to be accessed and updated reliably
and consistently.
• Service : Manages a collection of related resources and presents their functionalities
to users and applications.
• Seffer is basically storage of resources and it provides services to the
authenticated clients. It is running program on a networked computer. Server
accepts requests from client and performs a service and responds to request.
Example is Apache server and IIS server.
• 'The complete interaction between server machine and client machine, from the
point when the client sends its request to when it receives the server's response is
caUei a remote invocation.
• Reeources may be encapsulated as objects and accessed by client objects. In this
caee a client object invokes a method upon a server object.

• ~ of hardware resources that can be usefully be shored and "ample, of

IIMd.-e and softWlre resource sharing

tieirtharing.
....... MOUrcet:
t. CPU: tes processor-int£'n.c;ive applications for clients.
.. Computing Server : It execu
l,. lemote object server : It executes methods on behalf of clients.
cPV capacity of desktop machine with the local
Worm prograll\ : 1t shares

J
TECHNI~ PUBLICATIONS•• An"" lhMI lot /in(1Wllld9e
 & II -, Clcllt .. ,. hoJdl NCll\tly-
1cct1HCl web pages in its RA
It.,. -·· "' . .
1 Diak : N t I u 1-, ,ktl-ll1dlee ,lllp\dffi•
M,

lk 11Mr, video on demand server

&. 1 z ■ : Nd--. window aytllml, IUCh 11 X-11,
.
allow processes in ren,
cJ 11 f JI I ID . . . tlw, cm
wd ·of windows. .
S. p p p • N--►• pdnllll
acs:ept print jobs from many co
11 rs set 1 • mputers a
wlll\ a qllll1ing IJllem.
81 ........ :
L- . . : Wilt I l 81A ._,.11\Uldple clients to
sh are read-only page cont
H I -. .• • •'11114'111,ala d
l, ia tp share read-write files
.
- - • ~ IN lim
.- ld ll ll ll la
itl4ss. Shared whitebo
'1\ot • mmples of this type.
hl:Grd the definitive state of so
me re
. _ lllnce 'multi-user compute
rs appe
m t updates.
1DUel read-only copies of th
tlaoupout the Internet.
.. ._ an dlak and deliver the
 r f
f ig. 1.16.5

Solution : Lampart clocks ·

2 <5 ~tnelC b -+ h,
3 ,,,. 4 but c -, g. :,3
P2
P1
0
(1 oo11 • a
(2 0 0) 2 t b

(3 0 0 I 3 • c p3 3)4 ~ ;
:; • .. [3 .! 3) 3 t I [ "' , 3]
(AOOJ 4 • d
I '
' '
Fig. 1.1 6.6
AU : M.19-15
IIM Physical Clock Synchronization : NTP
• Clock synchroruzation is done by two me~'1ods :
1. Internal synchroruzation
2. External synchronization

external synchronization
• External synchronization refers to synchronization of process clocks C 1 \Vith an
authontative external source S) Let D>O be the synchronization bound and S be
the source of CTC. Then lS{t) - Ci(t, <D for i = 1, 2, ... , N and for all real times t.
• We say that clocks C I are accurate -Y.ithin the bound of D
TECHNJC.A.L PUBLJCJ.T,O,JS - Jin up tr.~J.!1 f u r ~ ~
 I

ti JUI•••• clocJcl C1 with each

,._. W Ct and c1 111 clock, at

.,,. . . . - llOt ftNlllldlY atlmllly

--of ,ro••,
-,., -1t .._, Ill dltft c:olllctlvtly
ti -,ndlronlztd
~ - · - • wlthlft bound 20.

:. C(t) that ii, • c1odc

dock ti.at
Atlock'1cruh
-_,flltan ii

l,oundaare

MPIL loYltr and upper

. . . m. 1ht ~
;-._.._totrlblmltm.
 ...,,_~Jllllll•t orJJ If tbl ~ II within I cert.1ri

a.& w, 311 • _,,_. ,err type alg rlth

-..aAf?~ ....... la~,-t"'_,. ., . o. m.
• 9- ..,,., ,ipdll' far _.,. t1w t1JM woWd be to ,unply JOsue a ••rno~
,...--...... - _.. and obWII the tline· .
• M all• n-- ta a a,ld,lnt with WWV rea,1ver, which receives P•ea,,
1

_, ..
IJIC.lt1C911d•--••--
t. A Moa _ . a nqi_.t to tlw
~

_..
07
d ta dia a,PffllUDI
tlJn! IJll5Vf!l al leul every d/2r 8eCOndi
~ aJJoWed ~ a clock and the UTC. '
a. n,, . . ••• _ . a nply -se with the CUJTl!RI UTC when receives U.
Nfld
& 'DI acJH •••--- the 1i1ne delay l,elWeell tiDle serve's ,sending u.
a P r md dia mzddM's Jlll!l1:iei,rin the _,,age. Then, ii uses the measure ~

., , .. dd-
~ - - . . . the cumnt ttDle from a tune ,erver.

lollToandT1 ............ wllhlhel8ffl8doCk

1o T
Cllfll---~-------~1 ----

--------------
........ Time_.

t.ff.t
Tt AN meuured using the same
.,_ • the ,ender's clock is off from

.. cr.o+Ti)/2
1
_.. •nr plus the tune tha
 afftklS aB ~ -
- - - - - ,rill be ..-vaiJable. It

b tbne· The tune at each

to flCCOIJllt for network

. . . . tllM (including its

. . time of
in

will be
• tblla of
- affllet
 t 4f

- N....- Time Protocol

• ~ • ~ , d M\\1 thl' U.r ktl• )
A~o rlth m "~ ln~ nde d for lntrnn'-?t •
8
• --flw Netwot\. TU M •~,.,vet
P'l'\,t\'X'\\l (NT I') ano thl! r met hod for syn<' hro nizl ng clocks
. . . \Ill a • I\M nl'd \k-. l arc hlt«
turv whe re the top level of the hicrMch
• . . . ~ to a UTC tim e 110\lrce 1uc. y
h R8 n GPS unit . •rt

- l.1a&11• AWflllnl Dlltrtbuttcl Atgortthm1

• lach aoclt exthanpe lta clock time wit h
its nel ghb on.

.. · ~
• 1""' _. hi dock time to the average of lte

. .
N IIL■ rk- ■ll6ol
ow n clock and the doc k times of its

,, :
l. 1111:h cl9 Ptl acrou the Int. ma t to be accurately syn
chr oni zed to UTC despite

..;z: : . .-Y lelt tha t can ,umve leng thy IOHN of con
nectivity.
C""an t k d'esfl lo lynchronl&e fNq uen tly and offN t the effe
cts of clock drif t.
6a PatrtOs "111DltCI\ ap iNt interference; auth
enti cate tha t the dat a is from a trusted
JI Pl
• 111111 NIP II vm ue COIIMClild In a logical hierarchy,
whe re servers in level n are
.l&sdly IO thole in level n-1 . The logical
hier arch y c.in be
ltlllr ■l bNDIM UNNChable or faile
d.
•••1•-• With one another In one of thre e mo des in the order 01

M N . . IN lm tt to be aynchronized
1a¥aNe . ,11. . . delaya encountered
 ~ T o ~ • milbl. s.n k, t~t can Aurvtv, lffl~thv l~'.H'fl of ronncdJ\.1ly
S. ~..._ clie nts to Mynchront e "Ulftdmtly, to ofket th!' rate of drift found
lftfflOlltl"OmpUtfta

'- To povict. inllrftrence with the time ur1t1 e, whcth r

protection •ma t
~ or acddental The time MrVk t utM 11uth#ntt tU.1 m te hruq es to
c:htck that tm\m8 data or;gtnatt from tN daimed tm~ted sour ~ It a~o
~ the ntum addl'flMt of fflffll8ft ,ent to u
• N'1P • • • synchronl&e modes an u follows :
1.. .... ...
I.~
S. s,ma .trkm ode.
i.•n at_ .:
• II M:Jll mode is intended for use on a high-speed LAN.
• Ole • &JR • vers periodically multicasts the time to the servers runrung m
....,.
. . . caarputers connectl!d by the LAN, which set their cloclcs assunung a small

but ones which nonetheless

ii 1111 aode mn only achieve relatively low accurate,
.., a rlf ed • ~ t for many purposes.
flu•• rn a• . . . :
.ftl u •me call mode is similar to the operation of Cristian's algorithm.
• :IJlllf eode. GIie Mter aa.-epts requests from other computers, which 1t pl'OC'eSSeS
tt•¥ fl,tc wlh • tia rrtamp-
ed
.... a1 :It ii Nls:NI whim higher accuracies are required that can be achiev
ar wt.e mullk:a,t is not supported in hardwue.
I! J Jfli ,. • • • on the NIM or a neighboring LAN, which need to keep
master sener m

-----
•tw ._.. thn for file a«r11•, c:ould contact a local

2111•:
.... II ••• drd for UN by the mute r eervers that supply time
ta U... 111d a,, 11w higher levela of the sync:h.roni.utian aubnet,
IC ;asAe • ID be achieved,
.... .... -,.m,tric mode exc han ge~ Warirlg timing

• put al an u1odation betweffl the eerven that is

Iha ll0CW"ICY of their 1ynchroruution over time,
 UNIT II

Me naa eO rde rln g

and Sna psh ots

■I JS t
I C a,I ·• ~ • q Jf I · • • · . . ~ ~Syncn ro,u)JlS
_• • • •,. . . ' ,. I ~·
'
•tlc aitA ~w:kt NIII r ..... .,,,..,. °"Oft DS\"ffi':M O!'IOllS

-Gt f C ) - r..,,,,/ . . . f(nf .. ,..,,,, . . , . Gl,fill __. elftll .,,.,,s111,t

■z ,, M I I f ~er ..W. J ft.8 WJP■ SMp:, W,,~h msfor FIFO

O P 3SMJ
• • • UI Al I -QJ s
II Ml 1S1 R ·•--S)slc...... oa ■ m•lc•lrJft

A a Q n r ••A11m 1R11a u1 ~-••

■ .. . . . . . ............... _....... · · · · · · · · . · . · · · ,,,~ 6
f4-•4~ •1111 itlal1Flaw

J Pmffllc.....,.-a,._,
a,RFOCA■ 1lll

,, • f}
 '\f 1,1,.. l~

~ 1hlr may be created and dt,troyed, P r ~ may JOffl (Jf

. - pre:e11■ may belong to mu1tip1, poupt,

pm¢111t1 to dta1 with coDectlont ol pr«-•• a, one abstraction
flOC•• lhould only lend a m••P to • group INS need not know oT
1191:la IIIIDlhera are.

--lllicatioft can be implemented in ,neral W"f',

2. Many to one 3. Many to 1111111

-1,y Cb MIider to multiple raiven.

ldw la alto known u J'llttJtic:Mt ~ A tpedal ca,e ol
- - - · .. bl'Oldcut COllllllUntcatlon.

,-0111• folllll a poup. ~ • ot two typa : CIOMd and

ie• 111 ...,_ of die poap Nlld a m•••P to the group.

C111
1{. . . . . . A mlN. to d,e pollJ) MI whoJe, although It
\¥"............. ot• gtoup.
• 1)1111: am Nlld • m ■r1p to the group as a

tw•-.. appllmllon.

'4 I I J

loallioll....,.....
. . . . . . . . . . . ti ..
. .tying .......
..... 111.,......
. . . . . . .-.i. . . . . . . .
. . . . be .,ndlranoul due to . . Wlutring

i .... pa«IIIIto wait until al thl -.twtng

. . . . . ..., - ready to JWIN dlf _, lkillt
 ...... _, mt N awaN of all IN _.Ylfts ~----~ tt,,.1

•
•·•·
•nNrret 1 ,ICHlfll' • not buttw.d for the ,_..tvln, pro, "lifi
- - • • proc111 II not tn a ••• ready to ,-,1v, It Thert!fc,r~, 11 ,tt
"'HI lb

.
aaus• a••·· I
■ I I ed GIiiy by 11'°91 proc11.,, of thtt multlc••• Aroup that ar~ rrncly

¥111411 .. bult.ffd for th• ~lvln,s

I proCl'IJ, I{) ,·,uh J•ffi(t ti
.r•
latl .. ••a*
II _.
··•1ttsc •two.,,.
....
•WJ•d paup wdl eventually ..-.tv• the m..1111•
WIida
of IIIIIAl\tkl for one--to-many communJcatlon,
• S Cla II 111 r ■ lla I A copy ol the mewp t, Hnt to each proci-11 of th~
■ •21t1 . . . and D'tlll . . . • buffertd until it t, accepted by tht proc~,,
I. 8 I 1 In sf t I •tlel a A • • • • to be multicat l, addlffHd
'1RQ11 di If••• 11111 ID, ffKJ lndlvtdual proc:NI of tM multicut group.
to a c.hnnnrl

t tlltl- la ■1111111 --• loatlon

I tfl Fl N ...... cliffeNnt depw of reliabWty.
--■■■ llkamlkP\. lht clepw of rellabWty t, normally exprc:••ll(•d 11,
1h ,, ' •
1 ltk e 1q 1h • II -,a1ld by the Nnder from any of the ttccivt•Ni
, Dia Jlwlll . . . . . a •pcne from any of the receiven.
t Ille mulltoeM poup COllltltl of n Neeiven and the ,endt·r
-•-·Q Can C n) of thl NCliven.
. . , . . I NlpGIIN lnllllp from all the recelven. l't

JEii • .,•
.. aurc ....... It II . . . received b~ All
,l...... ,,, . . it II not NCelvtKt t,\

- - plart
 I p

................ ,,..
IMaf wd and If any°"' .-in the 1et

11111" • deltvtnd to all remven

. , , . , IO updatll tht same record ol a

. . . ol lht datab•e. U tht metNg£
Ill dt&illll orden, then th! fin.al
- ba dlffeae,t in ill two replicas.

Rt R2 S2

lft1

,. S2
 :D:lstrli:,::bu::,:led::_:S~y_:st:em::':..__ _ _ _ _ _ _ _ _ _ 8_s_
2_._B _ _ _ _ _ _M sa
_r;_o_o_,_
c.11J_tli_nu_ ar=-
,a...,s_n;_;;a1,,t,,if
4

• Example : Oistributrd ~imulotion

• Dr•wbacks :
1. Too sttict con.,traint
2. No nbsolut<' ~ynchronircd clcxk
3. 'l' gunrnnt\;<.' ti, ratch nil trndy nwss,1g"s
Consistent I Total ordering
• h ensure:_- that all mt'ssagcs nrc delivered to all receiver processes in
order. the same
S1 R1 n; S2
• Fig. 2.4.3 show~ consistent ordering.
Ti I
• Rule : Messages received in the same
~rder, regardless of their timestamp.
• lmplement•tion : A message sent to a
sequencer, assigned a sequence number,
and finally multicast to receivers. A
message retrieved in incremental order at a
receiver. Fig. 2.4.3 Consistent orderfng
• Exunple : Replicated database updates
• Drawback : A centralized algorithm
Causal ordering
S1 R1 R2
• If two message sending events are not R3 S2
causally related, the two messages may
be delivered to the receivers in any
order.
• Two message sending events are said to
be causally related if they are correlated
by the happened before relation.
• Fig. 2.4.4 shows the causal ordering.
• Rule : Happened-before relation Fig, 2.4.4 Cauaal orderfng
k /
o If e\ e1i e h and k < I, then e i e e 1 1

If e,i = send(m) and e recef ve(m), then ei E e1,

o i:;;
1
o U e -+ e' and e' -+ e", then e ➔ e"
• Implementation : Use of a vector menage.
• Example : Distributed f1Je 11yatem

•• Drawbacks : 2. Broadcut 111wned

l, Vector as an overhead
 __
matnx taggmg-tn--cm1""11'11;-,grc:,:i;.,......_ __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
AU : May-lC)
Ill Total Order
re deln ered to nll rece1, er processes in the sazne
• It en...~res thnt all tn ;ag
order
• If l\ con-ect p ~ d hn.! m c-. gc m before it delh ers m', then any other correct
, dch, ts ; w-ill deli, er m before m'.
• F1g lcb.1 '-!,, rons1~tent ordcnng.
S1 R1 R2 S2

Fig. 2.6.1 Consistent ordering

• For each pair of processes Pi and Pi and for each pair of messages ~1 x and M,
that are delivered to both the processes, Pi is delivered l\t, before t--1} if and only
if P1 is delivered Mx before My
0 Rule: Messages received in the same order, regardless of their timestamp.
0 Implanenution : A message sent to a sequencer, assigned a sequence number,
and finally multicast to receivers. A message retrieved in incremental order at a
~ver.
c Example: Replicated database updates
• Drawback : A centralized algorithm

TECHNICAL PUBLICATIONS • An up thtu$1 for l l n o ~

~-~.;...-----------...::.2..:_·.:_11:.._ __~_ _ _M_es_s_sg_e_O_rd_e_rlng~a_nd_S_n_s_psh_ots
- ThrM Phase Distributed Algorlthm
s-ter:
• Phase 1 : Process multicast the mcssngc M with a locally unique tag and the local
time stamp to the group members.
• ~ 2 : The sender process awaits a reply from all the group metnbe1s who
respond with a tentative proposal for a revised timestamp for that message.
• Phase 3 : The process multicasts the final timestamp to the group in line.
lleceivers:
• Phase 1 : Receiver receives the message with a tentative timestamp. It updates the
variable priority.
• ~ 2: The receiver sends the revised timestamp back to the sender.
• Fmal timestamp is received from the multi-caster.
Ill Id Pr
••• _ ff..llrff
 p,ot IIHI -Nk, I ~ 1<"./>-t,f, ,

.. died• pafonned by a t1it-: ~ 1

. ..,. nor a Joe-al kemet an bf.

. . odur;m hat io hr
• • ol ~ met•age

a,,oiw, IOCf!lf maced 1e60U:~ ,

..,_...,s . . . . . . in - aitial ~
.......... l«tioJ ..,
to an 1/0 dmce

.........
.... ILJMd ,ylteM where.·

•ct.bibutedsy,tem.
fllllf!dlinaa:g die critical &ection
illlilllllnerufion

. , ..

me,sage

' .......
 o,ti.mpc,ritiprldun
f.wO .,,. ol ffldllP' ! REQUt91
- ~A

• lolloeJ dock whkh it updated lf!(urdJ, ~ b

- • - of awqu,tt,. ~ueffl are u,d~t"-j
IJIShfflP'; if dfflltWT1P9 are equal, j'llef.et

-
O ~ dw requnt ~a~ tu all utl~1
.,..,., Ptoe1111 it allowed to ~ntn lf!f! c~
• • •, ..., 111, reque,t tnn811ge (Ull5i3lc; ;A
t11ak)andit,~;
. . CS : awll••t requnted, ur hel<l

•dbl, it broadcast time,tamped

I may be tn one of three statts

11db\, Nl1d reply (OK) to fJCllder
reply and add the reque,t w •

ha ,ent its requcsl In

ltWld ...- p with the one that
, it sends a reply
lt II the winner and doc,

1111111111! JllPLY mettage from aU

 /J/811/lml• d M11/1JXancl ()ttlltlf",;k

,ttt St f>Ottl th, cs,

tiMtRl.
II llt•1,,h, ,m, p/V,I!(() ""'"'JflKIJII lo ,,11 11itC!t in its
tlte SI ~lVet1 th,, l{Bl.HASL(i) ll lll'lld" fl t<t!PLY(J)
11w10111w1 Il'lllll t11!1,,
the nC?xt t1lh_l Wallin~ 111 tlw
q11p111• 111111 d1•l1•1t·<1 tlwt 1•11try from th c
, the queue is tmpty, lhl'II lhti 1111,, upd,1h•11 ll11 ,.,ntf! to n:fli•ct thnt the
not k.-nt llUt nny NBPLY tnt'Ht4,tgP,

.:- . lite Si has r«elved REPLY J\ll'~Mgl• from n Hih• Sj, tlw uutho_riz:ition
ft
~ t in this message remainfl VAiid lmtll ~l !ll'nd~ n REPLY 11u:t1!l,1gc to SJ,
11• S,4.1 shows operation of Rkart-Agrawnln nlgorllhm .
• J"'O'
, • sue 51 and S2 are making request for criticnl IIC'Ction.
- 1.

Fig. 3.4.1 (a)

i lle 52 enters the critical section.

'-~enters
crt11catuctlon

,... U.1 (ia»

 f

nn n
\
53 '. sz 1,avos
cr,tlcnl soctlon

Fig, 3.4.1 (c)

Step 4 : Site 51 enters the critical section

S1 enters ,
critical section ' \
I

rl
s,nn n n Fig. 3.4.1 (d)

Review Question

1. Write short note on the following : Ricart and Agrawnla's algorithm.

AU : Mav-16. M&!~• I

IIJ Maekawa's Algorithm

• The main idea of the algorithm is to Jet the process that want to enter the cr1tic1·
section to compete for votes. Every process P has a voting district Sp H ,;
required that for all i and j, Si and SJ have at lcacit one common clcrnent.
• When a process wishes to enter the critical tiection, it sends a vote rcq\1 C!t t<> e11ef'J'
member of its voting district. When the processor receives rcpHe, from aU ~
. . l ·t ccefvtJ
members of the d 1str1c , 1 can enter the critical section. When n prcxe;.,501 r
-------~~~:-- -----------___..,
TECHNl~l PUSLICl<TIONS , An up fhrwt /er ~ I J
 3 • 11
-' got/! request, it respo d Dlstrtbutad Mutex ,ind oeadtock

n s criti
w· h
~voe
tfld' or other
a processor the
cand'd
. . exits itca1asectioi
· . . tf
"YES" vote . il has not • 1ready ca,twhich
its vote.
ates ,, it mforms tl,e voting
· district,
· can
t f 1

~E Pi Vi is.• chosen
• 1 Pi process Pi
A 1s ass
su oc1ated
. : • a votint set Vt of processes. The set Vi for the
. ch that with
. . process is in its ownv ·
2, Vi/"\ Vi
voMS ~ts
* ·{ J . The . oting ocess
set. m the overlap between any two
. re is at least one pr .

e e same size
3. ~.11 voting sets ar th
IVi I : I Vj I:
4. Eac process Pi is conta·med within . M . ·
• WheJ'\ h
aie,nbetS
a processor wants
of its district It mt o enter a cntic · voting
· al section,
.
sets. . sends a request to all
ti
· t ay
processor receives a reque
()n exit it informs its d' u:
.t enter' if it .gets a grant fr otn all members. When a
1 answers wi
th yes, if it has not already cast its vote.
ct to enable a new voting,
15
• As 1,efore each process ,naintainS a state variable which can be one of the

1- Released : Does not have access to the critical section and does not reqUlle ,t.
following:
2. wanted : J)oeS not have access to the critical section but does require it.

• S. }leld : currently
Ill addition haS access to the critical section.
each process ,naintainS a boolean variable indicating whether or not
the pfOCt!SS haS "voted". Of course voting is not a one-time action. This variable
really ;ndicateS whether some process within the voting set has access to th•
(litical section and haS yet to release it. To begin with, thes• variables are set to

~ " and False respectively.

· 1, Question
J. Wrift s/Wrl nolt on ~ fallaVJi"g : ~ • • vofing algorithm. ff Mi/ · :11

suzukl-Kasaml'•
. . the access
9,oadcast Algortthrn
right to the sh ared resource is passed in a
i/!I ,;Lotpral token represer'tiJ>S ~~-· whoever holds the token is allowed to

,,..tlld
~ eritieal ~
fas)iiol'sectiorL
l
the pro---~
· all . ,
,n,Ol!S 51teS•
.,..;.
.1.1....,.
means a sequence nuinbet is u..sed
 P2P and Distributed
Shared Memory

Introduction - Data indexing and <J\lerlays - Chord - Content addressable networks - Tapestry.
""'1fbldetl shared mmwry: Ab1traction and advantages - Memory consistency models -Shared
,nawny Mutual Exclusion.

Contents
8
5-1 Introduction of Peer-to-Peer Computing .... . . May-16, 17, 18, Dec.-17, · ·· Marks
5.2 Data Indexing and Overlays .......... . .. .. May-16, 17, 18, 19,
................ . . Dec.-16, 18 · · · · · · · · · · · Marks 13

5.3 Chord
5.4 Content Addressable Network
5.5 Tapestry
· \Jlll..6 Dlstrib·.,-"'
uun, Shared Memory . . . . . . . . . . . . . . . Dec.•18' · · · · · · · · · · · · · · Marks 8

5. 7 DSM Architecture
5.8 Design Issue In DSM
5.Q Memory Consistency Models
5.10 Shared Memory Mutual Exclusion
Two Marlcs Questions with Answers

(5 - 1)