Balanced RM2: An improved data placement scheme for tolerating double disk failures in disk arrays

Abstract

There is a growing demand for high data reliability beyond what current RAIDs can provide and users need various levels of data reliability. In order to meet these requirements, an efficient data placement scheme called RM2 has been proposed in [1], which enables a disk array system to tolerate double disk failures. However, RM2 has high data reconstruction overhead due to data dependency between parity groups in the case of double disk failures, suffering from large performance fluctuation according to disk failure patterns. To minimize the reconstruction overhead, this paper presents an improved data placement scheme called Balanced RM2. Experimental results show that the performance fluctuation of the Balanced RM2 becomes much less than that of the RM2 regardless of read or write. It is also shown that Balanced RM2 provides better performance than RM2 by 5 similar to 20% in the case of double disk failures.