On adaptive distributed storage systems based on functional MSR code

Abstract

Distributed storage systems (DSSs) provide a mechanism to store data reliably. There are broadly two different implementations of a DSS: replication based DSS and erasure code based DSS. For the same reliability, an erasure code based DSS needs to store significantly lesser amount of data. In recent times, the so called repair problem in an erasure code based DSS has found immense attention. There have been various proposals for a DSS considering the repair problem. One such proposal is the minimum storage regenerating (MSR) code based DSS. For a file M of size |M|, an (n, k, |M|,d, t) functional MSR code based DSS requires to download the minimum amount of data from d helper nodes to repair any t failed nodes. The parameters n, k, d, and t essentially determine the reliability of the DSS and the amount of data that needs to be downloaded to repair failed nodes. For a given reliability, there is a associated storage cost. We may not require the same reliability and/or there could be different constraints at different times on the number of helper nodes. Thus, it is desirable that a DSS implementation, such as a functional MSR code based DSS, should be capable of adapting to different values of the parameters n, k, d, and t in the most economical way. A coding scheme has been proposed in [14] which requires the minimum download while converting an (n 1 ,k 1 , |M|,d 1 ,t 1 ) functional MSR code based DSS into an (n 2 , k 2 , |M|,d 2 ,t 2 ) functional MSR code based DSS. In the present work, we generalize the work of [14].