摘要: Online services hosted in data centers show significant diurnal variation load levels. Thus, there is potential for saving power by powering down excess servers during the troughs. However, while techniques like VM migration can consolidate computational load, storage state has always been elephant room preventing this down. Migrating not a practical way to I/O load. This paper presents Sierra, power-proportional distributed subsystem centers. Sierra allows of large fraction troughs without migrating and imposing extra capacity requirements. It addresses challenges maintaining read write availability, no performance degradation, consistency, fault tolerance general workloads through set including power-aware layout, virtual log, recovery techniques, predictive gear scheduling. Replaying live traces from large, real service (Hotmail) on cluster shows savings 23%. Savings 40--50% are possible with more complex optimizations.
acm.org
sci-hub.se 参考文章(21)
Tony Guidici, Tejaswi Redkar, Windows Azure Platform ,(2010)
Fred B. Schneider, Robbert van Renesse, Chain replication for supporting high throughput and availability operating systems design and implementation. pp. 7- 7 ,(2004)
Gregory R. Ganger, Matthew Wachs, Michael Abd-El-Malek, Eno Thereska, Argon: performance insulation for shared storage servers file and storage technologies. pp. 5- 5 ,(2007)
Leonidas Rigas, Feng Zhao, Suman Nath, Lin Xiao, Wenbo He, Jie Liu, Gong Chen, Energy-aware server provisioning and load dispatching for connection-intensive internet services networked systems design and implementation. pp. 337- 350 ,(2008)
Christian Limpach, Eric Jul, Andrew Warfield, Steven Hand, Ian Pratt, Christopher Clark, Jacob Gorm Hansen, Keir Fraser, Live migration of virtual machines networked systems design and implementation. pp. 273- 286 ,(2005) , 10.5555/1251203.1251223
Brian M. Oki, Barbara H. Liskov, Viewstamped Replication: A New Primary Copy Method to Support Highly-Available Distributed Systems principles of distributed computing. ,vol. 62, pp. 8- 17 ,(1988) , 10.1145/62546.62549
Gregory R. Ganger, Shafeeq Sinnamohideen, Raja R. Sambasivan, John D. Strunk, Brandon Salmon, William V. Courtright, Jay J. Wylie, Andrew J. Klosterman, James Hendricks, Matthew Wachs, Michael Abd-El-Malek, Manish Prasad, Michael Mesnier, Chuck Cranor, Eno Thereska, Ursa minor: versatile cluster-based storage file and storage technologies. pp. 5- 5 ,(2005) , 10.5555/1251028.1251033
John R. Douceur, William J. Bolosky, Progress-based regulation of low-importance processes symposium on operating systems principles. ,vol. 34, pp. 247- 260 ,(1999) , 10.1145/319151.319168
Hrishikesh Amur, James Cipar, Varun Gupta, Gregory R. Ganger, Michael A. Kozuch, Karsten Schwan, Robust and flexible power-proportional storage Proceedings of the 1st ACM symposium on Cloud computing - SoCC '10. pp. 217- 228 ,(2010) , 10.1145/1807128.1807164
Haifeng Yu, Phillip B. Gibbons, Optimal inter-object correlation when replicating for availability principles of distributed computing. pp. 254- 263 ,(2007) , 10.1145/1281100.1281137