C 2007 by Qingbo Zhu. All Rights Reserved. PERFORMANCE AWARE ENERGY EFFICIENT STORAGE SYSTEMS

C 2007 by Qingbo Zhu. All Rights Reserved. PERFORMANCE AWARE ENERGY EFFICIENT STORAGE SYSTEMS

View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Illinois Digital Environment for Access to Learning and Scholarship Repository c 2007 by Qingbo Zhu. All rights reserved. PERFORMANCE AWARE ENERGY EFFICIENT STORAGE SYSTEMS BY QINGBO ZHU B.S., Nanjing University, China, 1999 M.E., Nanjing University, China, 2002 DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Computer Science in the Graduate College of the University of Illinois at Urbana-Champaign, 2007 Urbana, Illinois Abstract Trends in Internet infrastructure are driving towards using data centers to provide services such as web hosting and outsourced storage. Data centers typically have high power require- ments and may require as much electricity as an entire city. This makes energy a growing consideration in the Total Cost of Ownership for data centers. Among all components , disk storage is one of the biggest energy consumers. For a typical medium-sized data center, disk drives would represent an electricity budget of $7–9 million per year. Therefore, it is very important to reduce disk energy consumption in data centers. Designing an energy-efficient high-end storage system is a challenging task. Conserving energy usually incurs performance penalty; however, performance is very critical in data cen- ters as they must meet response-time Service Level Agreement they provide to customers. Moreover, compared to single disk systems, high-end storage systems consisting of hundreds or thousands of disks is much more complex. All components including storage cache man- agement, disk power control and data layout need to be carefully designed to minimize disk energy consumption in a performance-conscious manner. In this report, we present the design of a performance-aware energy-efficient storage system in data centers. We achieve this by combining a number of novel techniques: power- aware storage cache replacement algorithms and write polices that can provide more oppor- tunities for underlying disk power management to save energy; a coarse-grain global control algorithm that optimally determines the power configuration for each disk to minimize energy consumption based on the specified performance goal; a self-adaptive, performance-directed and energy-efficient data layout which dynamically balances between performance and en- iii ergy conservation as workloads change; a performance guarantee method that automatically boosts performance if there is a risk that performance goals might not be met due to disk power management. Our simulation results using both file system and online transaction processing(OLTP) traces show that our system can provide up to 65% energy savings while still meeting the response-time performance requirements (6.5–26 times better than previous solutions). Our OLTP emulated system results show that our system can save more energy (29%) than previous solutions, while still providing an OLTP transaction rate comparable to a RAID5 array with no energy management. We also study the interaction of different storage components in our system and how new storage alternatives affect the design of energy-efficient storage systems. iv To my mother v Acknowledgments My first and foremost thank goes to my adviser, Prof. Yuanyuan Zhou, for her guidance, support, and encouragement in the past five years. She has always been my major source of inspiration and motivation. She taught me not only how to conduct good research, but also how to be successful in all aspects of my life. I could not make this far without her. I would like to thank the other members of my dissertation committee, Prof. Sarita Adva, Prof. Klara Nahrstedt, and Prof. Sam King for their invaluable comments and constructive criticisms that greatly improved my thesis. I also thank Dr. Pei Cao at Google, Dr. Windsor Hsu at IBM Almaden Research Center, Dr. Kimberly Keeton and Dr. John Wilkes at HP labs with whom I happily collaborated with. They offered tremendous help in my research. Their insightful suggestions base on real industrial experience also made my work more complete, practical and relevant. Thanks also go to all members in the Opera research group. Special thanks to Zhifeng Chen, Zhenmin Li, Asim Shankar, Vivek Pandey, Francis David, Xiao Ma and Lin Tan who worked together with me in several projects in my Ph.D years. They offered me delightful discussion and unremitting assistance. I would also like to thank Pin Zhou, Feng Qin, Joseph Tucek, Shan Lu, Weihang Jiang, Soyean Park and Spiros Xanthos. It was so lucky to have them around, making the past five years a happy research journey. Finally, I would like to thank my dear mother for her unconditional love and support. She has always been there for me no matter what happened. I love you, mom! This thesis is dedicated to you. vi Table of Contents ListofTables..................................... ix ListofFigures..................................... x Chapter1 Introduction.............................. 1 1.1Motivation..................................... 1 1.2OurContributions................................ 5 1.3ReportOutline.................................. 6 Chapter2 BackgroundandRelatedWork................... 7 2.1DiskPowerModels................................ 7 2.2EnergyControlAlgorithms............................ 8 2.3StorageCacheManagement........................... 11 2.4DiskArrayLayouts................................ 12 2.5PerformanceGuarantees............................. 14 Chapter3 Power-AwareStorageCacheManagement............. 15 3.1Power-AwareOff-lineCachingAlgorithms................... 15 3.1.1 Energy-OptimalProblem......................... 15 3.1.2 Energy-OptimalAlgorithm........................ 17 3.1.3 Off-linePower-AwareGreedyAlgorithm................ 20 3.2Power-AwareOn-lineCachingAlgorithms................... 22 3.2.1 TheFirstAlgorithm:PA-LRU...................... 22 3.2.2 TheSecondAlgorithm:PB-LRU.................... 26 3.3Power-AwareWritePolices............................ 31 Chapter4 Coarse-GrainResponseEnergyControlAlgorithm....... 34 4.1Overview...................................... 34 4.2ProblemFormalization.............................. 35 4.3 Solving for Response Time Rij .......................... 36 4.4 Solving for Energy Eij .............................. 38 4.5FindingaSolution................................ 39 vii Chapter 5 Performance-Directed Energy-Efficient Adaptive Layout . 40 5.1Overview...................................... 40 5.2ChoosingtheDiskConfiguration........................ 41 5.3EfficientSmall-ScaleReorganization....................... 42 5.4EfficientLarge-ScaleReorganization....................... 44 5.5ThePerformanceModelRevisited........................ 48 Chapter6 ResponseTimePerformanceGuarantees............. 50 Chapter7 EvaluationMethodology....................... 51 7.1Trace-drivenSimulation............................. 51 7.1.1 TestBed.................................. 51 7.1.2 Traces................................... 54 7.2EmulatedSystemEvaluation........................... 56 Chapter8 SimulationResults........................... 58 8.1OverallResults.................................. 58 8.1.1 ResultsofStorageCacheManagement................. 58 8.1.2 Results of Control Algorithms, Layouts and Performance Guarantees 66 8.2SensitivityAnalysis................................ 70 8.2.1 EffectsofEpochLength......................... 71 8.2.2 EffectsofDataShuffling......................... 72 8.2.3 EffectsofResponseTimeGoals..................... 72 Chapter9 EmulatedSystemResults...................... 74 Chapter10 InteractionsofStorageComponents ............... 77 10.1EffectsofDiskSpin-up/downCosts....................... 77 10.2EffectsofMulti-SpeedPowerModel....................... 79 10.3EffectsofDiskArrayLayouts.......................... 81 10.4ImpactsonCPUandMemoryEnergy...................... 83 Chapter11 ImplicationsofNewStorageAlternatives............ 85 11.1IsSolidStateDriveaGoodStorageAlternative?............... 85 11.2APracticalApproach-HybridDrives...................... 89 Chapter12 ConclusionsandFutureWork................... 91 References....................................... 93 Author’sBiography..................................100 viii List of Tables 7.1 Simulation parameters for multi-speed disk model I ................ 52 7.2 Simulation parameters for multi-speed disk model II ................ 52 7.3Theevaluatedschemes.............................. 53 7.4 Default Synthetic Trace Parameters ........................ 55 8.1 The cost of data shuffling with and without foreground OLTP workloads . 72 11.1 Comparison between Flash Memory and Battery backed DRAM [Bit04] ...... 86 ix List of Figures 1.1Thearchitectureofamodernstoragesystem.................. 3 2.1 5-mode disk power model and lower envelope LE(t)function......... 9 3.1AnexampleshowingthatBelady’salgorithmisnotenergy-optimal...... 17 3.2Energysavingsovermode0foreachmodeina5-modedisk......... 23 3.3Approximatedcumulativedistributionfunctionofintervallengths....... 24 3.4AnexampleofenergyestimateinPB-LRUforadisk............. 30 4.1 Observed inter-arrival time distributions of OLTP and Cello99 workloads . 37 4.2Themeasuredandmodeledaveragediskresponsetime(OLTPtrace).... 39 5.1AdaptiveLayout................................. 42 5.2Large-scalereorganizationsolutions:PSandSS................ 45 5.3RandomizedShuffling............................... 46 7.1Theemulatedsystem............................... 56 8.1EnergyConsumption(normalizedtoLRU)..................

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    110 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us