论文部分内容阅读
在现代的多副本节能存储系统中,主副本的磁盘总是保持着开启以对外提供服务,而其他副本的磁盘则可以在负载低时保持关闭状态以节能。然而,因为新的写请求不能够被立即同步到所有的磁盘上,所以系统的可靠性被降低了。本文中,我们设计了一个基于RAID10的具有高可靠性和高能效的副本存储系统——RERAID。RERAID使用主副本磁盘上的一部分空闲空间结合纠删码构造了一个编码缓存来接收新的写请求。因为通过使用纠删码,编码缓存能够支持2个甚至更多磁盘的故障恢复,所以RERAID能够保证与RAID10存储系统想当的可靠性。另外,我们还设计了一个纠删码写(erasure coding write,ECW)算法来缓存那些小的随机写,并把它们合并成大的写数据块,然后并行写入编码缓存中以提高写性能。实验结果显示RERAID相比现有的方案能够提高写性能并且节省更多的能源。
In modern, multi-copy, energy-efficient storage systems, the primary replica’s disks are always left open for service while the other replicas are left powered down when the load is low. However, system reliability is reduced because new write requests can not be synchronized to all disks at once. In this article, we’ve designed a RAID10-based replication storage system, RERAID, with high reliability and energy efficiency. RERAID uses a portion of the free space on the primary copy disk in conjunction with the erasure code to construct a coded cache to receive new write requests. Because by using erasure codes, the code cache can support the recovery of two or more disks, RERAID ensures the reliability of RAID10 storage systems. In addition, we have also designed an erasure coding (ECW) algorithm to cache small random writes and merge them into large blocks of write data, which are then written in parallel to the code cache to improve write performance. Experimental results show that RERAID can improve write performance and save more energy than existing solutions.