The main objective of this research is to improve the performance of I/O subsystems in parallel architectures. I/O subsystem design is an issue which is often ignored by parallel computer designers. With the fast advances in VLSI and multiprocessor technology the processing power and main memory speeds have been increasing at rapid rates but the I/O system has hardly seen any comparable improvements. This disparity in performance gives rise to the I/O bottleneck in parallel architectures. The proposed investigation focuses on issues related to the improvement of I/O subsystem performance. Emphasis will be placed on I/O schemes for hypercube multiprocessor architectures. Efficient parallel I/O schemes for hypercube architectures are studied. Simulations are done mainly using some I/O bound problems, viz., external sorting, database manipulations, etc. Parallel algorithms are developed for such problems on the hypercube with a parallel I/O environment. Furthermore, different disk organizations and data distribution schemes for the parallel I/O systems will also be evaluated. The results obtained in this project will enable researchers to design database and transaction processing systems and provide insight for improving the performance of I/O subsystems on other architectures.
