On repairing erasure coded data in an active-passive mixed storage network

摘要

Networked distributed storage systems (NDSS) use erasure codes in lieu of replication for realising data redundancy. An interesting research challenge is to derive the largest advantage from the trade-offs between storage overhead and reliability that erasure codes provide, while optimising them to satisfy specific storage needs like repairability and better degraded read performance. Many coding strategies for NDSS exploit the storage nodes' computational ability to improve on repairability by applying network coding techniques. However, not every storage node in a NDSS is necessarily endowed with computing capability. This paper studies the effects of passive nodes, i.e., nodes without computational ability, on the repairability of erasure coded data, and how they may impair the promised performance of novel coding techniques. Specifically, we determine a lower bound on the minimum storage overhead in an active-passive mixed storage network, indicative of the price to pay in storage to achieve bandwidth efficient repairability.