Low bandwidth repair of the RS(10,4) Reed-Solomon code

摘要

As an alternative to replication of data blocks, the Hadoop Distributed File System offers the possibility of erasure coding using Reed-Solomon codes. The use of Reed-Solomon codes significantly reduces storage overhead but has more expensive failure recovery. Using the shortened Reed-Solomon code RS(10,4), with 10 data symbols and 4 check symbols, standard erasure repair requires downloading 10 symbols or 80 bits. Known schemes attain a reduced repair bandwidth of 65 or 64 bits. In this paper we present three repair schemes with bandwidth 60, 56 and 54, respectively.