Effect of task allocation methods for mesh-connected parallel computers considering task size limitation and machine size extension
Kuniyasu Suzaki, Hitoshi Tanuma, S. Hirano, Yuuji Ichisugi, M. Tukamoto · 2002
We analyzed the behavior of task allocation methods "Two Dimensional Buddy", "Adaptive Scan" and time sharing systems that use them for mesh-connected parallel computers under task size limitation and machine size extension. The efficiency of task allocation methods for mesh-connected parallel computers is expected to be impressed by the shapes and sizes of rectangular processor partitions which are required by tasks and a machine. In general the task allocation methods are expected to improve packing ability of tasks if the number of processor for a task is limited. However the execution time of the task is expanded. We examined performance trade-off between increased packing ability and task size limitation under task size scalability. Furthermore, we examined how to extend a parallel computer efficiently. From simulation results we confirmed that task size limitation improved performance of task allocation methods and that rectangular extension is better than square extension.