University of Houston

Storage class memories (SCMs) constitute an emerging class of non-volatile storage devices that promise to be significantly more reliable than magnetic disks. While RAID architectures have achieved a great success by introducing file striping, mirroring, and parity checking, they still cannot recover from more than two simultaneous disk failures. Even though SCMs promise to be much more reliable than magnetic disks, RAID solutions still lead in being the least expensive.

We propose to improve the reliability of conventional RAID level 5 arrays by adding one SCM device to each group of two or three RAID arrays and store on that device additional parity data. Our objective is to leverage the much higher reliability of the SCM devices while keeping the total system cost low by using a single SCM device to protect data stored on ten to twenty disk drives.

We show that the new organization can tolerate without data loss (a) all double disk failures, (b) between 75 and 90 percent of all triple disk failures and (c) between 50 and 70 percent of all failures involving two disks and the SCM device. As a result, the SCM parity device increases the mean time to data loss of the arrays in the group it protects by at least 20,000 percent.