Specification and management of QoS in real-time databases s(8)
The95percentconfidenceintervalsofatearelessthan½ateÀ2:1%;ateþ2:1% .ForBaseline-1andBaseline-2,theateincreasesastheloadincreases,violatingthereference,ater,atloadsexceeding175percent.InthecaseofPC-ATEEDF,atereachesthereferenceat150percentof
5.Wehavenotobservedanydeadlinemisses.
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
AMIRIJOOET
AL.:SPECIFICATIONANDMANAGEMENTOFQOSINREAL-TIMEDATABASESSUPPORTINGIMPRECISECOMPUTATIONS315
Fig.5.Experiment2:Averageperformancewhenvaryingexecutiontimeestimationerror.(a)PC-MPUandPC-MP.(b)PC-ATE.
appliedload.ForPC-ATEHEFandPC-ATEHEDF,atereachesthereferenceat175percent.AllPC-ATEalgorithmsprovidearobustperformancesinceateiskeptatthespecifiedreferenceduringoverloads.
Consideringsdte,the95percentconfidenceintervalsforallalgorithmsarelessthan½sdteÀ1:92%;sdteþ1:92% .Forallalgorithms,sdteincreasesasloadandateincrease.At200percentload,thecorrespondingsdteforPC-ATEEDF,PC-ATEHEDF,andPC-ATEHEFis34.1percent,25.2percent,and18.2percent,respectively.Consequently,deviationoftransactionerrorisminimizedwhenHEFschedulingisused.Itisworthmentioningthat,underHEFscheduling,sdteislessasresourcesareallocatedwithregardtotransactionerrors,while,underEDFscheduling,resourcesaredistributedwithregardtodeadlines.UnderEDFscheduling,twotransactionswithdeadlinesneareachothermayreceivedifferentamountsofresourcesand,hence,theymayterminatewithasignificantdeviationintransactionerror,while,inthesamecase,underHEFscheduling,theresourcesaredividedsuchthatbothtransactionsterminatewithtransactionerrorsclosetoeachother.
The95percentconfidenceintervalsofmdearelessthan½mdeÀ0:13%;mdeþ0:13% .StartingwithFig.4a,theaveragemdeforBaseline-1andBaseline-2violatesthereferencemdesetto2percent.Incontrast,inthecaseofPC-MPU,mdeissignificantlylowerthanmder.Sincethemisspercentagesarekeptlowatalltimes,theyarenotlikelytoovershoot.Consequently,thecontrolsignalfromthemisspercentagecontrollersislikelytobepositive,whichisinterpretedbytheQoDmanagerasaQoDupgradeand,hence,mdewillnotreachthelevelofmder.ThisisfurtherexplainedinSection5.6,wherethetransientperformanceofthealgorithmsisdiscussed.PC-MPpro-videsanaveragemdeclosertomder,givenby1:78Æ0:024%atloads150percentand200percent.However,mdedoesnotreachmdersincemdeissetaccordingtomO(whichdoesnotreachmOr).TurningtoFig.4b,the95percentconfidenceintervalsofmdearelessthan½mdeÀ0:5%;mdeþ0:5% .TheaveragemdeforBaseline-1andBaseline-2violatesthereferencemdesetto5percentatappliedloadsof175percentand130percent,respectively.Incontrast,inthecaseofPC-ATEEDF,PC-ATEHEF,andPC-ATEHEDF,mdeisatthereferenceduringoverloads.
TheexperimentshaveshownthatPC-MPUproducesamisspercentagesignificantlylowerthanthespecifiedmisspercentagesand,hence,itdoesnotfullysatisfythegivenQoSspecification.PC-MP,ontheotherhand,producesmisspercentagesclosetothegivenreferencesand,consequently,thegivenQoSspecificationissatisfiedwithregardtosteady-stateperformance.WehavealsoseenthatPC-ATEHEF,PC-ATEHEDF,andPC-ATEEDFarerobustagainstvaryingappliedloads.Moreover,PC-ATEHEFoutperformstheotheralgorithmswithregardtoQoSfairnessofadmittedtransactions.
5.5Experiment2:ResultsofVaryingesterr
Weapply200percentloadandvarytheexecutiontimeestimationerroraccordingtoesterr=0.00,0.25,0.50,0.75,and1.00.ControllerstunedusingSTAareused.Fig.5aandFig.5bshowtheperformanceofPC-MPU,PC-MP,PC-ATEHEF,PC-ATEHEDF,andPC-ATEEDF.Dash-dottedlinesindicatereferences.
AsinExperiment1,mMiszeroforallapproachesandesterr.TurningtomO,the95percentconfidenceintervalsforallalgorithmsarelessthan½mOÀ2:7%;mOþ2:7% .Asexpected,Baseline-1andBaseline-2donotsatisfytheQoSspecification,whereasPC-MPUandPC-MPareinsensitiveagainstvaryingesterr,asmOandmdedonotchangeconsiderablywhenvaryingesterr.Studyingate,wenotethatBaseline-1andBaseline-2donotsatisfytheQoSspecificationasatereaches52percentwhenesterrequals1.The95percentconfidenceintervalsforPC-ATEalgo-rithmsarelessthan½ateÀ0:3%;ateþ0:3% .PC-ATEEDF,PC-ATEHEF,andPC-ATEHEDFareinsensitiveagainstvaryingesterrasateandmdedonotchangewithvaryingesterr.Fromtheabove,wecanconcludethatPC-MPU,PC-MP,PC-ATEHEF,andPC-ATEHEDFareinsensitivetochangestoexecutiontimeestimationand,hence,theycaneasilyadaptwhenaccurateruntimeestimatesarenotknown.
5.6Experiment3:TransientPerformance
Studyingtheaverageperformanceisoftennotenoughwhendealingwithdynamicsystemsand,therefore,westudythetransientperformanceofPC-MPU,PC-MP,PC-ATEHEF,PC-ATEHEDF,andPC-ATEEDF.Wesetloadto200percentandesterrto1.
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
316IEEETRANSACTIONSONCOMPUTERS,VOL.55,NO.3,MARCH
2006
Fig.6.Experiment3:Transientperformance.(a)PC-MPU.(b)PC-MP.(c)PC-ATEEDF.
TABLE1
JsandJawith95PercentConfidence
Intervals
5.6.1ResultsofControllersTunedUsingSTA
Fig.6a,Fig.6b,andFig.6cshowthetransientbehaviorofPC-MPU,PC-MP,andPC-ATEEDFwithcontrollerstunedusingSTA.WerefertoTable1forasummaryoftheperformanceofPC-ATEHEF,PC-ATEHEDF,andPC-ATEEDFwithcontrollertunedusingSTAandDYN.Thedash-dottedlineindicatesthereference,whilethedottedlineindicatesmaximumovershoot.Forallalgorithms,mOandateovershootsdecayfasterthan60s,whichislessthanthesettlingtimerequirementgivenintheQoSspecification.StartingwithPC-MPU,wecannotethatmOiskeptlowatalltimes.ThisisexpectedsincetheaveragemOwasshowntobelow.Thereadermayhavenoticedthatmdeisgreaterthanzerointheinterval20-150,wheremOiszero.Sincemdeisgreaterthanzero,itisclearthat lERmaybecomenegativeduringthatpe …… 此处隐藏:6280字,全部文档内容请下载后查看。喜欢就下载吧 ……
相关推荐:
- [教育文库]夜场KTV服务员的岗位职责及工作流程[1]
- [教育文库]企划、网络、市场绩效考核方案
- [教育文库]学党史、知党情、强党性--“党的基本理
- [教育文库]2016年高考物理大一轮总复习(江苏专版
- [教育文库]干部廉洁自律自查自纠的报告
- [教育文库]2010年北京大学心理学系拟录取硕士研究
- [教育文库]资金时间价值练习题及答案
- [教育文库]保护环境的心得体会
- [教育文库]英语角内容:英语趣味小知识
- [教育文库]档案收集与管理工作通知
- [教育文库]劳动规章制度范本范本
- [教育文库]高考物理一轮复习课后限时作业1运动的
- [教育文库]机械工艺夹具毕业设计195推动架设计说
- [教育文库]通用技术教学比赛说课稿2
- [教育文库]2018年四年级英语下册 Module 7 Unit 2
- [教育文库]第2章 宽带IP网络的体系结构
- [教育文库]九年级化学第五单元课题3《根据化学方
- [教育文库]小学英语六年级情态动词用法归纳
- [教育文库]甲级单位编制窑井盖项目可行性报告(立
- [教育文库]2016-2021年中国城市规划行业全景调研
- 高考英语听力十大场景词汇总结
- 全省领导班子思想政治建设座谈会会议精
- 人教版新课标高一英语提优竞赛试题 下
- 江西省2014年生物中考试题
- 长沙镇食品药品安全事故应急预案
- 《金刚石、石墨和C60》片段教学设计
- 福州教育学院(王旭东)
- 基于EDA音乐播放器的设计
- 9、古诗两首《夜书所见》《九月九日忆
- 小学语文课外阅读有效策略探讨
- 贵州文化产业发展成支柱产业的问卷调查
- 膀胱类癌的诊治体会(附3例报告)
- 发动机积碳产生的原因
- Configuring Code Composer Studio for
- 学生良好的心理素质如何培养点滴谈
- 46 电沉积法制备锂离子电池用硅-锂薄膜
- 美舍雅阁公司管理中各部门职责
- 去壳剥皮的小妙招
- 六自由度运动平台的仿真研究
- Pride and Prejudice(傲慢与偏见)




