Efficient data mappings for parity-declustered data layouts

AbstractThe joint demands of high performance and fault tolerance in a large array of disks can be satisfied by a parity-declustered data layout. Such a data layout is generated by partitioning the data on the disks into stripes and choosing a part of each stripe to hold redundant information. Thus the data layout can be represented as a table of stripes. The data mapping problem is the problem of translating a data address into a disk identifier and an offset on that disk. Recent work has yielded mappings that compute disks and offsets directly from data addresses without the need to store tables. In this paper, we show that parity-declustered data layouts based on commutative rings yield mappings with improved computational efficiency and wider applicability