Ripple-RAID:一种面向连续数据存储的高效能盘阵

视频监控、备份、归档等应用具有独特的负载特性和I/O访问模式,需研究特定的存储节能方法.磁盘阵列的局部并行策略有利于实现该类存储系统的节能,但通常会导致RAID执行小写操作而严重影响性能.为此,提出一种面向该类存储系统的高效能盘阵——Ripple-RAID,采用新的局部并行数据布局,通过综合运用地址转换、异地更新、基于流水技术渐进生成校验、分段数据恢复等策略,在单盘容错条件下,保持了局部并行的节能性,又有效解决了局部并行带来的小写问题.Ripple-RAID具有突出的性能和节能效率,在80%顺序写负载情况下,请求长度为512KB时,写性能为S-RAID 5的3.9倍,Hibernator、MAID写性能的1.9倍,PARAID、eRAID 5写性能的0.49倍;而比S-RAID 5节能20%,比Hibernator、MAID节能33%,比eRAID 5节能70%,比PARAID节能72%.

Ripple-RAID: A High-Performance and Energy-Efficient RAID for Continuous Data Storage

The applications, such as video surveillance, backup and archiving, have inherent workload characteristic and I/O access pattern, and require specialized optimization for storage energy saving. Partial parallelism in RAID is beneficial to storage energy saving, but generally makes RAID perform small writes, which heavily deteriorates the performance. A high-performance and energy-efficient RAID, Ripple-RAID, is proposed for these applications. A new partial-parallel data layout is presented, and by comprehensively employing strategies such as address mapping, out-place update, generating parity data progressively based on pipeline, and segmented data recovery, Ripple-RAID not only obtains energy efficiency of partial parallelism but also eliminates the small writes incurred by partial parallelism while providing single disk fault tolerance. When write workload is 80% sequential and transfer request size is 512KB, the write performance of Ripple-RAID is 3.9 times that of S-RAID 5, 1.9 times that of Hibernator and MAID, and 0.49 times that of PARAID and eRAID 5; meanwhile its energy consumption is 20% less than S-RAID 5, 33% less than Hibernator and MAID, 70% less than eRAID 5, and 72% less than PARAID.