Оценка размера рабочего набора виртуальной машины исходя из гостевых показателей производительности (1187411), страница 8
Текст из файла (страница 8)
×òî íå óäèâèòåëüíî, òàê êàê â ðàìêàõ äàííîé ìîäåëè ñèñòåìà íå ìîæåò îòäàâàòü ïîä ôàéëîâûé êåøâñå ñâîáîäíûå ðåñóðñû.TestWindows 7, 1CPU, 4Gb Windows 8, 1CPU,4Gbbusyloop_test−2.0%−2.0%system_syscall_test−0.1%−0.3%process_exec_test−2.0%−2.3%thread_create_test−2.0%−1.8%io_hdd_seq_rand_rd_test−99.8%−99.2%virtalloc_test−1.1%−0.2%mem_read_test−1.5%−5.4%76mem_write_test−1.6%−2.9%mem_pf_read_test−0.2%−10.0%mem_pf_write_test+0.4%−9.1%mem_copy_test−1.2%−1.6%Òàáëèöà 7.1: Ðåçóëüòàòû òåñòèðîâàíèÿ ïðîñòîé îöåíêè ðàçìåðà ðàáî÷åãî íàáîðà èç ðàçäåëà 6.1.Äëÿ òîãî ÷òîáû ÎÑ Windows íå ïûòàëàñü èñïîëüçîâàòü ïîä ôàéëîâûå êåøè âñþ ñâîáîäíóþ ïàìÿòü, íóæíî ñäåëàòü ñëåäóþùåå:1.
Ïîñòàâèòü â ñîñòîÿíèå Enable LargeCacheFile â ðååñòðå Windowsäëÿ èçìåíåíèÿ ïîëèòèêè êîíòðîëÿ êåøà[41];2. Óñòàíîâèòü ãðàíèöû ðàçìåðà êåøà ïðè ïîìîùè ñèñòåìíûõ âûçîâîâ[26].Äëÿ òåñòîâ ðàçìåð êåøà áûë óñòàíîâëåí â ãðàíèöàõ îò 256Mb äî768Mb.Ðåçóëüòàòû òåñòîâ ñ èçìåíåííîé ïîëèòèêîé ïðåäñòàâëåííû â òàáëèöå 7.2. Êàê ìîæíî óâèäåòü èç òàáëèöû âêëþ÷åíèå êîíòðîëÿ êåøàóëó÷øàåò ðåçóëüòàò, íî âñå æå äàííàÿ ìîäåëü íå ïîäõîäèò äëÿ èñïîëüçîâàíèÿ â ñèñòåìàõ, àêòèâíî ðàáîòàþùèõ ñ ôàéëàìè.77TestWindows7, Windows7,compared to a VM compared to thewith no memory VM with memorymanagementmanagementWindows 7withoutcontrolbusyloop_test+1.1%+3.2%system_syscall_test−1.0%−0.9%process_exec_test−3.3%−1.4%thread_create_test+3.1%+5.0%io_hdd_seq_rand_rd_test−99.8%+32.0%virtalloc_test−0.4%+0.7%mem_read_test−2.4%−0.9%mem_write_test−2.4%−0.8%mem_pf_read_test−0.3%−0.0%mem_pf_write_test−0.1%−0.6%mem_copy_test−0.9%+0.3%Òàáëèöà 7.2: Ñðàâíåíèÿ ðåçóëüòàòîâ òåñòèðîâàíèÿ ïðîñòîé ìîäåëè èç ðàçäåëà 6.1 ñ êîíòðîëåì êåøà è áåç íåãî.78cache7.2Òåñòèðîâàíèå ìîäåëè íà îñíîâå âûäåëåíèÿ ïàòòåðíîâÏðîòîòèï ìåòîäà èç ðàçäåëà 6.2 áûë ðåàëèçîâàí â Matlab 2014.Ïîëó÷åííûå äàííûå î ïîòðåáëåíèè ïàìÿòè áûëè ðàçáèòû íà íåïåðåñåêàþùèåñÿ íàáîðû îáó÷àþùèõ è òåñòîâûõ.
Íàïðèìåð, åñëè âåëèñüíàáëþäåíèÿ â òå÷åíèè 2-õ äíåé, òî ïåðâûé äåíü èñïîëüçîâàëñÿ êàêèñòî÷íèê äëÿ îáó÷àþùèõ äàííûõ, âòîðîé äåíü êàê èñòî÷íèê äëÿ òåñòîâûõ. Òåñòèðîâàíèå ïîñòðîåííîãî àëãîðèòìà ãîâîðèò î òîì, ÷òîàëãîðèòì îøèáàåòñÿ â 6.4% ñëó÷àåâ.79ÃËÀÂÀ 8Âûâîäû äàííîé ðàáîòå ïðèâåäåíî 2 àëãîðèòìà îöåíêè ðàçìåðà ðàáî÷åãîíàáîðà âèðòóàëüíîé ìàøèíû.Äàëüíåéøåå íàïðàâëåíèå èññëåäîâàíèÿ íàïðàâëåííî íà âñòðàèâàíèå ìîäåëè èç ðàçäåëà 6.2 â ñèñòåìó Parallels Desktop è ïðîâåäåíèåòåñòîâ.Îäíèì èç âîçìîæíûõ íàïðàâëåíèé èññëåäîâàíèÿ ÿâëÿåòñÿ ïîïûòêà àâòîìàòè÷åñêîãî îáíàðóæåíèÿ ñòðóêòóðíûõ èçìåíåíèé áîëååïðîñòûìè ìåòîäàìè, íàïðèìåð, êëàññèôèêàöèÿ âðåìåííîãî ðÿäà íàèìåþùèé ñòðóêòóðíîå èçìåíåíèå è íå èìåþùèé.80ÃËÀÂÀ 9Çàêëþ÷åíèåÖåëü äàííîé ìàãèñòåðñêîé äèññåðòàöèè ðàçðàáîòàòü àëãîðèòì,êîòîðûé îöåíèâàåò ðàçìåð ðàáî÷åãî íàáîðà âèðòóàëüíîé ìàøèíû èïðèíèìàåò ðåøåíèå î òîì, ñêîëüêî îïåðàòèâíîé ïàìÿòè ìîæíî èçúÿòü ó êîíêðåòíîé âèðòóàëüíîé ìàøèíû(virtual machine, VM) äëÿ ïåðåðàñïðåäåëåíèÿ ìåæäó äðóãèìè VM, çàïóùåííûìè íà òîé æå õîñòìàøèíå (host machine). ðàáîòå áûë ïðîâåäåí àíàëèç ïîòðåáëåíèÿ âèðòóàëüíîé ìàøèíîé ïàìÿòè ñ ñòàòèñòè÷åñêîé òî÷êè çðåíèÿ.
Áûëî ïîêàçàíî, ÷òî ðåãðåññèè îòperformance countersèìåþò íèçêóþ ïðåäñêàçàòåëüíóþñïîñîáíîñòü. Àâòîðåãðåññèè òàê æå ïîêàçàëè ïëîõîé ðåçóëüòàò. Ðÿäûÿâëÿþòñÿ íåñòàöèîíàðíûìè è ñ íàëè÷èåì ñòðóêòóðíûõ èçìåíåíèé,÷òî ãîâîðèò î òîì, ÷òî òàêèå ìåòîäû, êàê àâòîðåãðåññèè, ôèëüòðÊàëüìàíà è ò.ä. íå ðàáîòîñïîñîáíû.Ïðåäëîæåííî 2 âàðèàíòà îöåíêè ðàçìåðà ðàáî÷åãî íàáîðà âèðòóàëüíîé ìàøèíû.
Ïåðâàÿ ìîäåëü ïîêàçàëà õîðîøèé ðåçóëüòàò, íîíåïðèãîäíà äëÿ èñïîëüçîâàíèÿ â ñèñòåìàõ, êîòîðûå àêòèâíî ðàáîòàþò ñ ôàéëîâîé ñèñòåìîé. Ïðîòîòèï âòîðîé ìîäåëè äàâàë ëîæíûåïðåäñêàçàíèÿ â 6.4% ñëó÷àåâ.Ðåçóëüòàòû èññëåäîâàíèé ïî äàííîé ðàáîòå îïóáëèêîâàííû â [22,9].81Ëèòåðàòóðà[1] Gerald J Popek and Robert P Goldberg. Formal requirements forvirtualizable third generation architectures.ACM, 17(7):412421, 1974.Communications of the[2] Peter J Denning. The working set model for program behavior.Communications of the ACM, 11(5):323333, 1968.[3] ScottDLowe.Bestpracticesforoversubscriptionofcpu, memory and storage in vsphere virtual environments.VMware's White paper, available at: https://communities.vmware.com/servlet/JiveServlet/previewBody/21181-102-128328/vsphereoversubscription-best-practices [1].
pdf, 2013.[4] John J Prevost, KranthiManoj Nagothu, Brian Kelley, andMo Jamshidi. Prediction of cloud data center networks loads usingSystem of Systems Engineering(SoSE), 2011 6th International Conference on, pages 276281.stochastic and neural models. InIEEE, 2011.[5] Patrick Thibodeau. Data centers are the new polluters@ONLINE,2015. URL http://www.computerworld.com/article/2598562/data-center/data-centers-are-the-new-polluters.html.[6] Truong Vinh Truong Duy, Yukinori Sato, and Yasushi Inoguchi.Performance evaluation of a green scheduling algorithm for energyParallel & Distributed Processing,Workshops and Phd Forum (IPDPSW), 2010 IEEE InternationalSymposium on, pages 18.
IEEE, 2010.savings in cloud computing. In82[7] Anna Melekhova. Machine learning in virtualization: Estimate aCloud Computing (CLOUD),2013 IEEE Sixth International Conference on, pages 863870.virtual machine's working set size. InIEEE, 2013.[8] Carl A Waldspurger. Memory resource management in vmware esxserver.ACM SIGOPS Operating Systems Review, 36(SI):181194,2002.[9] Melekhova A. and Markeeva L. Estimating working set size byguest os performance counters means. CLOUD COMPUTING 2015,page 48, 2015.[10] Sheng Di, Derrick Kondo, and Walfredo Cirne. Host load predictionProceedingsof the International Conference on High Performance Computing,Networking, Storage and Analysis, page 21.
IEEE Computer Societyin a google compute cloud with a bayesian model. InPress, 2012.[11] Anderson Ravanello, Luis Villalpando, Jean-Marc Desharnais, AlainApril, and Abdelouahed Gherbi. Associating performance measureswith perceived end user performance.CLOUD COMPUTING 2015,page 125, 2015.[12] Zhenhuan Gong, Xiaohui Gu, and John Wilkes. Press: PredictiveNetwork and ServiceManagement (CNSM), 2010 International Conference on, pages 9elastic resource scaling for cloud systems.
In16. IEEE, 2010.[13] Kinshuk Govil, Dan Teodosiu, Yongqiang Huang, and Mendel83Rosenblum.Cellular disco: Resource management using virtualACM SIGOPSclusters on shared-memory multiprocessors.Operating Systems Review, 33(5):154169, 1999.[14] AndreaCArpaci-DusseauandRemziHArpaci-Dusseau.Information and control in gray-box systems. InACM SIGOPSOperating Systems Review, volume 35, pages 4356. ACM, 2001.[15] Pei Cao, Edward W Felten, and Kai Li.Implementation andProceedingsof the 1st USENIX conference on Operating Systems Design andImplementation, page 13. USENIX Association, 1994.performance of application-controlled le caching. In[16] Carl A Waldspurger and William E Weihl.
Lottery scheduling:Proceedingsof the 1st USENIX conference on Operating Systems Design andImplementation, page 1. USENIX Association, 1994.Flexible proportional-share resource management. In[17] Carl Waldspurger, William E Weihl, et al.An object-orientedframework for modular resource management.
In Object-Orientationin Operating Systems, 1996., Proceedings of the Fifth InternationalWorkshop on, pages 138143. IEEE, 1996.[18] Rudolph Emil Kalman. A new approach to linear ltering andprediction problems.Journal of Fluids Engineering, 82(1):3545,1960.[19] James H Stock and Mark W Watson.Introduction to econometrics,volume 104. Addison Wesley Boston, 2003.84[20] William Bell and Steven Hillmer. Initializing the kalman lter fornonstationary time series models.Journal of Time Series Analysis,12(4):283300, 1991.[21] Stefan Frey, Simon Disch, Christoph Reich, Martin Knahl, andNathan Clarke. Cloud storage prediction with neural networks.CLOUD COMPUTING 2015, page 67, 2015.[22] Ìåëåõîâà À.Ë. Ìàðêååâà Ë.Á.
Òîðìàñîâ À.Ã. Îäíîðîäíîñòü âèðòóàëèçàöèîííûõ ñîáûòèé, ïîðîæäåííûõ ðàçëè÷íûìè îïåðàöèîííûìè ñèñòåìàìè.Òðóäû ÌÔÒÈ, (1), 2014.[23] Sayanta Mallick, Gaetan Hains, and Cheikh Sadibou Deme.A resource prediction model for virtualization servers.InHigh Performance Computing and Simulation (HPCS), 2012International Conference on, pages 667671. IEEE, 2012.[24] US Air Force.
Analysis of the intel pentium's ability to support asecure virtual machine monitor. 2000.R 64 and ia-32 architectures software developer's[25] Part Guide. Intelmanual. 2010.MicrosoftWindows Internals: Microsoft Windows Server 2003, Windows XP,and Windows 2000, volume 4. Microsoft Press Redmond, 2005.[26] Mark E Russinovich, David A Solomon, and Jim Allchin.[27] James Bernard Ramsey. Tests for specication errors in classicalJournal of the RoyalStatistical Society.
Series B (Methodological), pages 350371, 1969.linear least-squares regression analysis.85[28] Íèêèòà Âÿ÷åñëàâîâè÷ Àðòàìîíîâ. Ââåäåíèå â ýêîíîìåòðèêó. ÍÂÀðòàìîíîâ.-: ÌÖÍÌÎ, 2011.[29] K Person. On lines and planes of closest t to system of points inspace. philiosophical magazine, 2, 559-572, 1901.[30] Halbert White. A heteroskedasticity-consistent covariance matrixestimator and a direct test for heteroskedasticity.Econometrica:Journal of the Econometric Society, pages 817838, 1980.[31] Stephen M Goldfeld and Richard E Quandt.homoscedasticity.Some tests forJournal of the American statistical Association,60(310):539547, 1965.[32] Ëáîâ Ã.Ñ.
Áåðèêîâ Â.Á.Ñîâðåìåííûå òåíäåíöèè â êëà-Âñåðîññèéñêèé êîíêóðñíûé îòáîð îáçîðíîàíàëèòè÷åñêèõ ñòàòåé ïî ïðèîðèòåòíîìó íàïðàâëåíèþ¾Èíôîðìàöèîííî-òåëåêîììóíèêàöèîííûå ñèñòåìû, 2008.ñòåðíîì àíàëèçå.[33] Andrea Vattani. K-means requires exponentially many iterationseven in the plane.Discrete & Computational Geometry, 45(4):596616, 2011.[34] Ethem Alpaydin.Introduction to machine learning.MIT press,2014.[35] Oded Maimon and Lior Rokach.Data mining and knowledgediscovery handbook, volume 2. Springer, 2005.[36] Chotirat Ann Ratanamahatana and Eamonn Keogh. Everything86you know about dynamic time warping is wrong. In Third Workshopon Mining Temporal and Sequential Data.
Citeseer, 2004.[37] Hiroaki Sakoe and Seibi Chiba. Dynamic programming algorithmoptimization for spoken word recognition.Acoustics, Speech andSignal Processing, IEEE Transactions on, 26(1):4349, 1978.[38] Fumitada Itakura. Minimum prediction residual principle applied tospeech recognition.Acoustics, Speech and Signal Processing, IEEETransactions on, 23(1):6772, 1975.[39] George Casella and Roger L Berger.Statistical inference, volume 2.Duxbury Pacic Grove, CA, 2002.[40] Anthony McCluskey and Abdul Ghaaliq Lalkhen.Central tendency and spread of data.Statistics ii:Continuing Education inAnaesthesia, Critical Care & Pain, 7(4):127130, 2007.[41] O. Pentakalos M. Friedman.Windows 2000 Performance Guide.2002.87.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
