Abstract: In order to avoid data loss due to hardware failure or server breakdown,redundancy coding technology is widely employed in distributed storage systems for data reliability.However,traditional erasure codes,such as Reed-Solomon codes,bear the burden of huge rebuilding data amount.Compared with the replication technique,which only needs to read and transfer the lost data,the erasure coding requires to read and transfer a much large amount of data,thereby consuming much more disk I/Os and network bandwidth.Thus,a erasure code based distributed storage system would cost longer time for data reconstruction than a replication based system,and exposes the whole system in a long-term degraded stage,increasing the risk of the permanent data loss.To solve this problem,many repair-bandwidth-efficient codes were constantly proposed,but these codes are only compared with the traditional Reed-Solomon codes and lack the comprehensive comparisons on practical storage systems.We systematically analyzed these repair-bandwidth-efficient codes from the some significant aspects,such as amount reduction on reconstruction data and so on,thus providing valuable basis and references for choosing suitable erasure codes for practical systems.

Key words: Erasure codes,Data construction,Storage system,Distributed system