An interleaved file system for a highly distributed MIMD architecture

A file system design with a large number of disks, each with its own local processing power, is proposed. This organization allows concurrent access to many records of a file, thus providing a speedup proportional to the number of disks. The focus is on the structure of a single file. Care is taken to preserve the file's integrity, as several processes are contending for it from different processing elements (PEs). A file structure, based on the B/sup link/-tree, a modification of the well-known B/sup +/ tree, is introduced. This structure is distributed among the disk-nodes, the PEs that are equipped with disks. Performance figures are derived by analytic and simulation methods.<
>