ebook img

The nom Profit-Maximizing Operating System PDF

67 Pages·2015·2 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview The nom Profit-Maximizing Operating System

5 1 0 2 The nom Profit-Maximizing Operating - 5 1 System - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n Shmuel (Muli) Ben-Yehuda e i c S r e t u p m o C - n o i n h c e T 5 1 0 2 - 5 1 - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n e i c S r e t u p m o C - n o i n h c e T 5 1 0 2 The nom Profit-Maximizing Operating - 5 1 System - 5 1 0 2 - C S M s i Research Thesis s e h T . c Submittedinpartialfulfillmentoftherequirements S . M forthedegreeofMasterofScienceinComputerScience - t n e m t r a p e D e c n Shmuel (Muli) Ben-Yehuda e i c S r e t u p m o C - n o i n h c e T SubmittedtotheSenate oftheTechnion—IsraelInstituteofTechnology Iyar5775 Haifa May2015 5 1 0 2 - 5 1 - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n e i c S r e t u p m o C - n o i n h c e T This research thesis was done under the supervision of Prof. Dan Tsafrir in the Computer ScienceDepartment. Someresultsinthisthesisaswellasresultsthisthesisbuildsonhavebeenpublishedasarticles 5 1 bytheauthorandresearchcollaboratorsinconferencesandjournalsduringthecourseofthe 0 2 author’smaster’sresearchperiod. Themostup-to-dateversionsofthesearticlesare: - 5 OrnaAgmonBen-Yehuda,MuliBen-Yehuda,AssafSchuster,andDanTsafrir. TheriseofRaaS:The 1 - Resource-as-a-Servicecloud. CommunicationsoftheACM(CACM),57(7):76–84,July2014. 5 1 NadavAmit,MuliBen-Yehuda,DanTsafrir,andAssafSchuster. vIOMMU:efficientIOMMUemulation. 0 InUSENIXAnnualTechnicalConference(ATC),2011. 2 - Orna Agmon Ben-Yehuda, Eyal Posener, Muli Ben-Yehuda, Assaf Schuster, and Ahuva Mu’alem. C Ginseng: Market-drivenmemoryallocation. InACM/USENIXInternationalConferenceonVirtual S ExecutionEnvironments(VEE).2014. M OrnaAgmonBen-Yehuda,MuliBen-Yehuda,AssafSchuster,andDanTsafrir. DeconstructingAmazon s EC2spotinstancepricing. ACMTransactionsonEconomicsandComputation(TEAC),1(3):16:1, si September2013. e h MuliBen-Yehuda,OmerPeleg,OrnaAgmonBen-Yehuda,IgorSmolyar,andDanTsafrir. Thenonkernel: T Akerneldesignedforthecloud. InAsiaPacificWorkshoponSystems(APSYS),2013. . AbelGordon,NadavAmit,NadavHar’El,MuliBen-Yehuda,AlexLandau,DanTsafrir,andAssaf c S Schuster. ELI: Bare-metal performance for I/O virtualization. In ACM Architectural Support for M. ProgrammingLanguages&OperatingSystems(ASPLOS),2012. MichaelHines,AbelGordon,MarcioSilva,DilmaDaSilva,KyungDongRyu,andMuliBen-Yehuda. - Applicationsknowbest:Performance-drivenmemoryovercommitwithGinkgo. InIEEEInternational t n ConferenceonCloudComputingTechnologyandScience(CloudCom),2011. e m t r a p e ACKNOWLEDGEMENTS D e First and foremost, I’d like to thank my amazing wife, friend, co-author, and advisor, Orna c n AgmonBen-Yehuda. Youtaughtmemorethanyouwilleverknow. Second,I’dliketothank e i c myamazingchildren,YaelandZe’ev,whomakeitallworthwhile. Third,I’dliketothankmy S parents, Yoel and Irit Ben Yehuda, for having kept faith all these years, even when my path r e t meandered. Last,I’dliketothankMichaelFactorandOrranKrieger,whotaughtmewhatit u p meanstodoresearch. m Thenomoperatingsystemandthisthesishavebeeninthemakingforalongtime. During o C theyearsIworkedonthem,Ipublishednearlytwentypapersco-authoredwithmanywonderful - people. I’dliketothankallofthem—ithasbeengreatworkingwithyou! n o i n h c e T ThegenerousfinancialsupportoftheTechnionisgratefullyacknowledged. 5 1 0 2 - 5 1 - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n e i c S r e t u p m o C - n o i n h c e T 5 1 0 2 Contents - 5 1 - 5 1 0 2 - C ListofFigures S M Abstract 1 s i s AbbreviationsandNotations 3 e h T 1 Introduction 5 . c S . 2 Motivation 9 M 2.1 Dynamicresourcepricingiscoming . . . . . . . . . . . . . . . . . . . . . . . 9 - t 2.2 Dynamicpricingmandateschange . . . . . . . . . . . . . . . . . . . . . . . . 10 n e m 3 Design 11 t r a 3.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 p e 3.2 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 D 3.3 CPUandscheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 e c 3.4 Memorymanagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 n e 3.5 I/Odevices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 i c S 3.6 Networking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 r 3.7 Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 e t u 3.8 Price-awareness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 p m 4 Economicmodelandutilityofnetworkbandwidth 17 o C - 5 Implementation 21 n o i 6 Evaluation 23 n h 6.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 c e T 6.2 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 6.3 Whatmakesnomfast? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 6.4 Profit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 6.5 Whatmakesnomprofitable? . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 6.6 Effectofbatchingonthroughputandlatency . . . . . . . . . . . . . . . . . . . 30 6.7 Throughput/latencyParetofrontier . . . . . . . . . . . . . . . . . . . . . . . . 31 7 Discussion 35 8 Relatedwork 37 5 1 9 Conclusionsandfuturework 39 0 2 - HebrewAbstract i 5 1 - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n e i c S r e t u p m o C - n o i n h c e T 5 1 0 2 List of Figures - 5 1 - 5 1 0 2 1.1 Cloudeconomicmodel: Applicationsruninthecloud. Userspaytheapplicationowner - C fortheservicetheapplicationprovides. Theapplicationownerinturnpaysthecloud S M providerforthecloudresourcestheapplicationuses(e.g.,networkbandwidth). . . . . 5 s i 3.1 Traditionalkernelstructurecomparedwithnom’skernelstructure. . . . . . . . . . . 12 s e h T 6.1 memcachedthroughputandlatency . . . . . . . . . . . . . . . . . . . . . . . . 25 . 6.2 nhttpdthroughputandlatency . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 c S 6.3 NetPIPEthroughputandlatency . . . . . . . . . . . . . . . . . . . . . . . . . . 26 . M 6.4 memcachedprofit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 - 6.5 nhttpdprofit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 t n e 6.6 NetPIPEprofit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 m 6.7 memcachedprofit: staticvs.adaptivebehavior . . . . . . . . . . . . . . . . . . . 29 t r a 6.8 nhttpdprofit: staticvs.adaptivebehavior . . . . . . . . . . . . . . . . . . . . . 29 p e 6.9 NetPIPEprofit: staticvs.adaptivebehavior . . . . . . . . . . . . . . . . . . . . 30 D 6.10 memcachedthroughput(inthemanyusersscenario)andlatency(inthesingleuser e c scenario)asafunctionofbatchingdelay . . . . . . . . . . . . . . . . . . . . . . . 31 n e 6.11 nhttpdthroughput(inthemanyusersscenario)andlatency(inthesingleusersce- i c S nario)asafunctionofbatchingdelay . . . . . . . . . . . . . . . . . . . . . . . . 31 r 6.12 NetPIPE throughput (in the many users scenario) and latency (in the single user e ut scenario)asafunctionofbatchingdelay . . . . . . . . . . . . . . . . . . . . . . . 32 p m 6.13 ThememcachedthroughputandlatencyParetofrontier . . . . . . . . . . . . . . . 33 o 6.14 ThenhttpdthroughputandlatencyParetofrontier . . . . . . . . . . . . . . . . . 33 C 6.15 TheNetPIPEthroughputandlatencyParetofrontier . . . . . . . . . . . . . . . . 33 - n o i n h c e T 5 1 0 2 - 5 1 - 5 1 0 2 - C S M s i s e h T . c S . M - t n e m t r a p e D e c n e i c S r e t u p m o C - n o i n h c e T

Description:
provider for the cloud resources the application uses (e.g., network bandwidth) . 5 —Aeschylus. 1. Technion More and more of the world's computing workloads run in virtual machines on Infrastructure- as-a-Service (IaaS)
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.