| Description: |
In this paper, we propose an affinity-based self scheduling scheme (ABS) for software shared memory system. In this scheme, the static affinity between processor and initial data distribution is considered when scheduling, and the synchronization overhead is reduced greatly when load imbalance occurs. Comparized with previous schemes, ABS performs better in metacomputing environment. 1 Introduction Loop scheduling has been extensively studied in past years, and many schemes were proposed, such as Self Scheduling(SS) [7], Block Self Sch eduling (BSS), Guided Self Scheduling (GSS)[5], Factoring Scheduling (FS) [2], Trapezoid Self Scheduling (TSS) [8], Affinity Scheduling (AFS) [4], Safe Self Scheduling (SSS) [3], Adaptive Affinity Scheduling (AAFS) [9], etc. However, all these previous work focused on either shared memory multiprocessors or dedicated hardware distributed shared memory systems, i.e., they assumed that the available computing power of each processor was equal, an. |