Efficient distributed fault-tolerant storage with erasure codes

We study the design of a distributed storage sys-tem that uses erasure codes to maintain data redun-dancy while disks can be written concurrently by multiple nodes. We propose and analyze a proto-col that is highly optimized for the common cases; in these cases, a read requires only a round-trip to the disk, and a write requires only a round-trip to each disk that requires updating. To tolerate tp node failures and td disk failures, the protocol uses = 1 + (tp + 1)(td + tp=2 1) redundant disks.