SGER: Investigating Degrading Failure Recovery in Large Scale, Heavily Utilized Disk Storage Systems

garth Gibson，卡内基梅隆大学首席研究员，研究小组：研究大规模、高利用率磁盘存储系统中降低故障恢复的问题。该研究小组的建议回应了NARA对文件系统和I/O研究的兴趣，特别是对基于超大磁盘系统的容错、高可扩展、持续I/O的方法和因素的研究。私家侦探曾在一次研讨会上预测过，在越来越大的磁盘存储系统中，磁盘大小不断增加会带来危险。为了更好地理解这一推测的含义，CMU将探索在磁盘尺寸增加的情况下，对大型磁盘存储系统的性能和可靠性的影响，导致故障恢复期间性能下降的时间增加，以及存储数据容量的扩展，导致可能失败的组件总数增加。P.I.和他的团队将探索不同的策略来权衡容量开销、性能开销和数据可靠性，例如，在更冗余的数据中更慢的恢复，在更多性能下降的情况下更快的恢复，以及在特定故障模式的点解决方案中使用冗余的新方法。

AbstractOCI 0852543Garth Gibson, Principal InvestigatorCarnegie Mellon UniversitySGER: Investigating Degrading Failure Recovery in Large Scale, Heavily Utilized Disk Storage SystemsThis SGER proposal responds to an interest from NARA in File Systems and I/O research, specifically, in methods and factors contributing to fault tolerant, highly scalable, sustained I/O for very large disk based systems. The P.I. had predicted in a workshop the dangers of increasing disk sizes in ever larger disk storage systems.In order to better understand the implication of this speculation, CMU will explore the implications on performance and reliability in large scale disk storage systems of the increasing size of magnetic disks, causing the periods of degraded performance during failure recovery to grow, and the scaling of stored data capacity, causing the total number of components that might fail to grow. The P.I. and his team will explore different strategies for trading off capacity overhead, performance overhead and data reliability, for example, slower recovery in more redundant data, faster recovery with more performance degradation, and novel uses of redundancy for point solutions of specific failure modes.