Systematic Mapping Of Reed-Solomon Erasure Codes On Heterogeneous Multicore Architectures

A classical concept of building fault-tolerant systems consists of replicating data on several servers. Erasure codes assume another approach which is based on splitting up the data into blocks, encoding them redundantly, and distributing the blocks over various servers. The use of erasure codes could radically improve the availability of distributed storage in comparison to replication systems with similar storage and bandwidth requirements. The investigation reported in this chapter confirms the advantage of using modern heterogeneous multi/many-core architectures, especially GPUs, for the efficient implementation of the Reed-Solomon erasure codes.