Scheduling loop applications in software distributed shared memory systems

This paper first examines the issues related to scheduling loop application s on a software distributed shared memory (DSM) system. Then, a dynamic sch eduling scheme is developed based on the examined issues to enhance the per formance of loop applications on DSM. Compared with previous works, the pro posed scheme has several specialties. The first is that the workload of pro cessors can be effectively balanced even when the computational capabilitie s of processors and the computational needs of threads are not identical. T he second is it divides thread mapping into two phases, each with one consi deration, i.e,, load balance or communication cost, and adopts thread migra tion and exchange in the two phases, respectively. The third is the exploit ation of data sharing among threads to reduce data-consistency communicatio n, and the last is to attack the negative effect of the unnecessary inter-n ode sharing caused by thread re-mapping. The proposed scheme has been imple mented on a page-based DSM system called Cohesion. Our experiments show tha t the proposed scheme is more effective to improve the performance of the t est programs than related schemes.