Specification and management of QoS in real-time databases s(9)

Abstract—Real-time applications such as e-commerce, flight control, chemical and nuclear control, and telecommunication are becoming increasingly sophisticated in their data needs, resulting in greater demands for real-time data services that are provided

AMIRIJOOETAL.:SPECIFICATIONANDMANAGEMENTOFQOSINREAL-TIMEDATABASESSUPPORTINGIMPRECISECOMPUTATIONS317

considerthedynamicsofthecontrolledsystem.Hence,itdoesnotconsiderthat,givenacertain lER,itmaytakeafewsamplesuntilthecorrespondingateisreached.There-fore,theSTAcontrollerpersistswithchangingtheloaduntilthedesiredaterisreached,atwhichpoint,ateovershootsduetoaggressivecontrol.ThisishandledmoreefficientlywiththeDYNcontrollerasitistunedaccordingtoadynamicmodeland,hence,thecontrollerismoregentlewhencontrollingthesystem.

Table1givesasummaryoftheperformanceofthecontrollerswithrespecttoJsandJa,whichshowhowcloselythecontrolledvariableatefollowsitsreferenceater.TheperformanceofPC-ATEEDFandPC-ATEHEDFissignificantlyimprovedwhenusingDYNfortuningcon-trollers.However,theperformanceimprovementforPC-ATEHEFistoosmalltobeconsideredsignificant.Fromtheresultsanddiscussionsinthissection,welearnedthattheschedulingpolicyofEDFinducesacertaindelaybetweenachangeintheloadandate,whereasthedelayislessfortheHEFschedulingpolicy.ThedelaycausedbyEDFiscompensatedforbytheDYNcontrollerasopposedtotheSTAcontroller,whichdoesnotcompensatefordelays,and,hence,weachievealmostthesameperfor-manceasPC-ATEHEF,whereaDYNcontrollerisused.TheexperimentsshowthatfeedbackcontrollersforPC-ATEEDFandPC-ATEHEDFtunedusingDYNoutperformcontrollerstunedusingSTA.

5.7SummaryofResultsandDiscussion

OurexperimentsshowthatPC-MPU,PC-MP,PC-ATEHEF,andPC-ATEHEDFareOrobustagainstinaccurateexecutiontimeestimationsasm,ate,andmderemainunaffectedforvaryingexecutiontimeestimationerrors.PC-MPUkeepsmOlessthanitsreferenceandisabletoefficientlysuppressdeadlinemisspercentageovershoots.PC-MPUshould,therefore,beappliedtoRTDBswheredeadlinemisspercentageovershootscannotbetolerated.PC-MPprovidesanmOnearitsreference,butgeneratesovershootsgreaterthanthemaximumallowedovershoot.TheexperimentsshowthatPC-MPisparticularlyusefulwhenmOmustbenearitsreference,butwhereovershootsareaccepted.ItwasobservedthatPC-ATEHEFprovidesalowersdtecomparedtootheralgorithms,loweringthedeviationoftransactionerroramongterminatedtransactions.ThispropertyisusefulinapplicationswhereQoSfairnessamongtransac-tionsisemphasized.Further,wesawthatthecontrolperformanceisgreatlyenhancedwhenusingDYNascomparedtoSTA[11].ThisaidsasystemoperatorinconfiguringRTDBsthatarehighlyreactivetochangesinappliedloadandexecutiontimeestimationerrors,provid-ingincreasedperformancereliabilityandenhancedQoSadaptation.

6RELATEDWORKLiuetal.[3]andHanssonetal.[25]presentedalgorithmsforminimizingthetotalerrorandtotalweightederrorofasetoftasks.Theirapproachesrequiretheknowledgeofaccurateprocessingtimesofthetasks,whichisoftennotavailableinRTDBs.BestavrosandNagyhavepresentedapproachesformanagingtheperformanceofRTDBs,wheretheexecutiontimeofthetransactionsisunknown[26].Eachtransactioncontributeswithaprofitwhencompletingsuccessfully.Anadmissioncontrollerisusedtomaximizetheprofitofthesystem.TheworkbyLiuetal.Hanssonetal.,andBestavrosandNagyfocusesonmaximizingorminimizingaperformancemetric(e.g.,profit).Thesepreviousapproachescannotbeappliedtoourproblemsince,inourcase,wewanttocontrolasetofperformancemetricssuchthattheyconvergetowardtheirreferencesasgivenbyaQoSspecification.

Luetal.havepresentedafeedbackcontrolschedulingframeworkwheretheyproposealgorithmsformanagingthemisspercentageand/orutilization[11].Incomparisontotheproposedalgorithmsinthispaper,theydonotaddresstheproblemofmaximizingQoSfairnessamongadmittedtasks.Further,theirmodelstaticallyrelatesestimatedrequestedload,deadlinemissratio,andutiliza-tion.Inthispaper,efeedbackcontrolschedulingtocontrolthelengthofaqueueofremoteprocedurecalls(RPCs)arrivingataserver[10].Incontrasttotheirworkwehavechosendeadlinemisspercentage,utilization,andaveragetransac-tionerrorasthecontrolledvariables.

efeedbackcontrolschedulingtomanagethedeadlinemissratiooftransactionsandfreshnessrequirementsofdataobjects[23].IncontrasttotheworkbyKangetal.,wehave,inthispaper,describedasetofalgorithmsformanagingQoSbasedonfeedbackcontrolschedulingandimprecisecomputation,whereQoSisdefinedintermsoftransactionanddatapreciseness.Further,wehaveintroducedQoSfairness,asetofnovelQoDmanagementalgorithmsincludingtwonewschedul-ingalgorithms(HEFandHEDF),andadynamicmodelgivingamoreaccuratedescriptionofthecontrolledsystem.KuoandHohaveintroducedthenotionofsimilarity[27],whereasimilarityrelationgiveswhethertwotransactionsproducesimilarresults.However,theworkbyKuoandHodoesnotaddressunknownworkloadcharacteristics.

DavidsonandWattersproposedamethodforgeneratingmonotonicallyimprovinganswersinRTDBsanddistrib-utedRTDBs[28].Aqueryprocessor,APPROXIMATE[7],producesanapproximateanswerifthereisnotenoughtimeavailable.Theaccuracyoftheimprovedanswerincreasesmonotonicallyasthecomputationtimeincreases.Therelationaldatabasesystemproposedin[29],canproduceapproximateanswerstoquerieswithincertaindeadlines.Leeetal.studiedtheperformanceofreal-timetransactionprocessinginbroadcastenvironments[30].Incontrasttotheapproachesabove,wehaveintroducedprecisionatthetransactionlevelandthedataobjectlevelandmanageQoSusingfeedbackcontrol.

7CONCLUSIONS

AND

FUTUREWORK

Inthispaper,wehavearguedfortheneedforincreasedadaptabilityofapplicationsthatprovidereal-timedataservices,whileoperatinginhighlyunpredictableenviron-ments.Typically,transactionscannotbesubjecttoexactschedulabilityanalysisgiventhelackofaprioriknowledgeoftheworkload,makingtransientoverloadsinevitable.

Abstract—Real-time applications such as e-commerce, flight control, chemical and nuclear control, and telecommunication are becoming increasingly sophisticated in their data needs, resulting in greater demands for real-time data services that are provided

318Furtherm …… 此处隐藏：6008字，全部文档内容请下载后查看。喜欢就下载吧 ……