高端磁盘阵列可靠性测试建模方法研究

阐述了衡量高端磁盘阵列可靠性的重要指标MTTDL,通过分析两种造成磁盘错误的原因,建模计算了各种RAID级别的MTTDL,设计了磁盘读出错率的测试方案,实现了基于分布式的磁盘读出错率测试程序。

MP-RAID: Mirroring and parity protected disk array

Redundant array of inexpensive disk （RAID）10 is known as the most reliable disk array architecture to tolerate up to half of total disk failures, but failure of two disks in the same mirror set still results in data loss. In this paper, we propose a new disk array architecture, mirroring and parity protected RAID （MP-RAID）, which combines both mirroring and parity techniques to further improve reliability of disk arrays. The main idea behind MP-RAID is to protect the data by both mirroring and parity techniques： keep two copies of data in the same mirror set and update the parity block in the log disk within the same parity groups. Reliability analysis shows that the reliability of MP-RAID, in terms of mean time to data loss （MTTDL）, is much better than RAID10 and RAID5.

非MDS码存储系统的通用可靠性模型

为了量化基于非最大距离可分码的分布式存储系统的可靠性,从非最大距离可分码的构造矩阵入手,提出了一种求解采用非最大距离可分码编码的数据对象在丢失若干块后数据对象的可修复概率算法。该算法穷举丢失若干块的所有可能组合,并在生成矩阵中判断每种组合相对应的矩阵是否可逆以计算可恢复的概率。随后采用马尔可夫理论,针对此类系统建立较为通用的度量存储系统可靠性的理论模型。该模型能够量化非最大距离可分码容错配置、存储规模、修复带宽、单节点可靠性、单节点容量对存储系统可靠性的影响。最后采用数值分析的方法,以局部修复码为例验证了模型的正确性,比较了不同因素对存储系统可靠性的影响。本模型为采用非最大距离可分码的存储系统的设计和实现提供了理论基础。