Efficient processor management strategies for multicomputer systems

Multicomputers are cost-effective alternatives to the conventional supercomputers. Contemporary processor management schemes tend to underutilize the processors and leave many of the processors in the system idle while jobs are waiting for execution;Instead of designing faster processors or interconnection networks, a substantial performance improvement can be obtained by implementing better processor management strategies. This dissertation studies the performance issues related to the processor management schemes and proposes several ways to enhance the multicomputer systems by means of processor management. The proposed schemes incorporate the concepts of size-reduction, non-contiguous allocation, as well as job migration. Job scheduling using a bypass-queue is also studied. All the proposed schemes are proven effective in improving the system performance via extensive simulations. Each proposed scheme has different implementation cost and constraints. In order to take advantage of these schemes, judicious selection of system parameters is important and is discussed