Improving DSS efficiency with shortened MSR codes

Abstract

The distributed storage system (DSS) is a network constructed by large number of nodes interconnected among themselves. The main issues in this area are the reconstruction process or obtaining the entire original message out of the DSS and the repair process or recovering the lost stored data of a failed node. The product matrix regenerating codes as class of codes in such networks offer an optimal tradeoff between the repair bandwidth and the data amount stored per node. In our paper, we are proposing a new code, a shortened MSR code, that gives new bound of the wanted tradeoff. The shortened MSR code is constructed by introducing modification on the mother code, product matrix MSR code. By comparing both constructions we prove that our code gives better performances regarding the repair bandwidth and the number of connected nodes needed for the repair and the reconstruction processes, while maintaining the storage per node unchanged. The comparison is extended between the new shortened MSR code and a new product matrix MSR code designed according to the parameter values obtained from our new code. The necessity of this is having a fair and convincing comparison results, which shows again that our code has some improvements in the overall distributed system.