Causes of deadlocks. Four necessary conditions for deadlock to occur are: The first three properties are generally desirable

Transcription

1 auss of dadloks Four ssary oditios for dadlok to our ar: Exlusiv ass: prosss rquir xlusiv ass to a rsour Wait whil hold: prosss hold o prviously aquird rsours whil waitig for additioal rsours No prmptio: a rsours aot prmptd from a pross without aortig th pross irular wait: thr is a st of lokd prosss ivolvd i a irular wait Th first thr proprtis ar grally dsiral rsptivly to i) prsrv rsour itgrity, ii) iras rsour utilizatio, iii) rdu wast of PU tim

2 Dadlok hadlig poliis Dadlok prvtio th systm is dsigd so that gratig rqusts vr lads to a dadlok Dadlok dttio th systm priodially (or wh dadlok susptd) hks for dadloks, ad a rovry produr is startd if o is dttd Dadlok avoida rsours ar gratd oly if th rsultig systm stat is saf i.. thr is at last o squ of xutio i whih all prosss ru to ompltio

3 akr s algorithm Dfiitios for a systm with prossors ad m rsours: ( ) a m a a matrix - vail Max m m m Max - laim matrix m m m urrt lloatio matrix ( ) k m k d d d D matrix vail urrt m m m E E E E urrt Nd matrix ( ) i m i i i f f f F tor Rqust v

4 lgorithm stps Stp - ttativ apt of rqust D : D - F // updat urrt vail matrix D i : i + F i // updat urrt llo vtor i E i : E i - F i // updat urrt Nd matrix E i Stp - saf-stat hkig tst s xt slid i whih frmoy D loa[i] i d[i] E i Stp 3 - if tst positiv dfiitiv apt of rqust, othrwis roll ak th updats of stp

5 akr at work: saf-stat hkig Whil (last_itratio_sussful) last_itratio_sussful fals ; for i to N do if (fiishdoutful[i] ND d[i] frmoy) th // d laim - loa, how // muh pross i still ds fiishdoutful[i] fals ; last_itratio_sussful tru fr moy fr moy + loa[i] ; // pross a fiish aus // d fr rsours // pross is al to fiish thus // it will rpay th loa ak d if d for Ed whil if (fr moy apital) th saf! ls ot saf! ;

6 akr s algorithm xampl () Systm with 3 prossors ad 3 rsours, ad with th followig matris: Rqust F ( ) should gratd? ( ) 3 4 matrix - vail Max - laim matrix Max lloatio matrix urrt ( ) k k D matrix urrt vail E urrt Nd matrix

7 akr s algorithm xampl () Suppos th rqust is gratd, th rsultig stat would th followig: lloatio matrix urrt ( ) k k D matrix urrt vail E urrt Nd matrix

8 akr s algorithm xampl (3) Th w stat is saf P 3 a omplt (aus E 3 D) ad thus a rtur ( ) to th pool of availal rsours D oms ( ); th outstadig ds of oth P ad P a satisfid

9 Dadlok aalysis Dadlok aalysis is ompliatd y th umr of variats of th pross-rsour rlatioship how may ad xatly whih rsours ar dd? How may tims a rsour a usd? How may prosss a simultaously us a rsour? W will study: simpl modls usd to rprsts rqusts ad rsour typs Irasigly sophistiatd graph-asd thiqus for dadlok aalysis

10 Modls of dadlok Dpdig o th typ of rsour rqust, a dadlok a lassifid aordig to o of four typs Sigl-uit rqust modl: vailla varity ND rqust modl: I d a PU ND a disk ND a pritr OR rqust modl: I d a disk OR a pritr ND-OR rqust modl: I d a PU ND (a disk OR a pritr)... P-out-of-Q rqust modl: I d at last thr ossus vots out of fiv

11 Wait-for-graph (WFG) To study dadloks th stat of a systm is oft rprstd with a wait-for graph p i p j p i ds a rsour hld y p j yl i th graph rprsts a irular wait

12 Dfiitio of kot kot of a graph is a sust K of ods suh that th rahal st of ah od i K is xatly K Exampls: D D,, ar i a yl ad also i a kot,, ar i a yl, ot i a kot

13 Modl diffrs Sigl-uit rqust modl: a dadlok orrspods to a yl i th WFG ND rqust modl: sam as aov (if vry rsour is i sigl opy), ut a pross ow a ivolvd i mor tha o dadlok OR rqust modl: a kot is a suffiit oditio for a dadlok ND-OR rqust modl: a kot is a suffiit oditio for a dadlok P-out-of-Q rqust modl: a kot is a suffiit oditio for a dadlok