On the I/O Costs in Repairing Short-Length Reed-Solomon Codes

Minimizing the repair bandwidth, i.e., the amount of information from the helper nodes needed for recovering the content of one failed node in an erasure-coded distributed storage system, has been the focus of many works in the literature. We investigate another important performance metric, namely the I/O cost, which specifies the amount of information that needs to be read by the helper nodes during the repair process of one failed node. We analyze the I/O costs of a few known repair schemes for Reed-Solomon codes of various lengths, in contrast to the previous works in this direction, which only studied the I/O costs in repairing full-length Reed-Solomon codes.