9303583 Scheuermann This is the first year funding of a three-year continuing award. Large arrays of small disks provide an attractive approach for high performance systems since they allow for low-cost, reliable storage and can achieve higher throughput compared to large disks. This project proposes to develop an integrated set of algorithms and an intelligent file manager which is applicable to any multi-disk architecture in which the operating system can access the disks independently. The emphasis of the approach is on the use of modular building blocks for file partitioning, allocation, load balancing, and reorganization that can be invoked independently of each other. Thus, the procedure for load balancing can be called regardless of whether striping or declustering is employed and does not depend upon the unit of declustering. The development of a generalized method for load balancing is considered which can be applied to an environment where some files exhibit periodical access patterns and which considers explicitly the cost of performing the cooling actions. The investigation also focuses on reorganization procedures that perform restriping when higher performance standards need to be enforced. The anticipated results of this project are of importance to massively parallel computers to support scalability and to ensure that the I/O is not the limiting factor in achieving high speedup. ***
