Exact Cooperative Regenerating Codes with Minimum-Repair-Bandwidth for Distributed Storage

Published in IEEE INFOCOM 2013

Abstract. Regenerating codes are a class of erasure codes for distributed storage systems that minimize the repair bandwidth when recovering a failed node. When multiple nodes fail simultaneously, cooperative regenerating codes allow the replacement nodes to cooperate during the repair process to further reduce the repair bandwidth. In this paper, we present constructions of exact cooperative regenerating codes that achieve the minimum repair bandwidth for all parameters in the cooperative model. Our codes are based on product matrix framework and interference alignment techniques, and they support exact repair (the repaired node is identical to the failed node) for any number of simultaneous failures. We analyze the storage overhead and repair bandwidth trade-offs and show that our constructions meet the fundamental bounds for cooperative regeneration.