Abstract
In parallel computers, an efficient allocation strategy is one of the most important task to optimize system utilization. A good processor allocation strategy is the one that efficiently selects a subset of the system such that the selected subsystem is the best one in terms of processor utilization. In this thesis, a good way of partitioning an arbitrary multicomputer system architecture is developed. Mesh, debruijn and hypercube structures are tested. Kernighan-Lin heuristic is used for partitioning algorithm. Various partitioning schemes are tested. Results for 2 and 4 partitions are presented in this thesis. Communication capabilities among the processors in this thesis. Communication capabilities among the processors in a subset, (ICCB), is considered as the measure of performance. Standard deviation of ICCBs of the same level are considered as the measure of goodness of partitioning for regular and symmetric architectures. A concept of relative ICCB or RICCB is introduced to compare different architectures. It is concluded that 2 partition gives better results for mesh and bypercube while 4 partition gives better results for debruijn graph.
Shahed, Sumon (1994). Partitioning an arbitrary architecture. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1994 -THESIS -S5257.