| Summary: | 碩士 === 靜宜大學 === 資訊管理學系研究所 === 95 === With the development of cheap personal computers and high-speed networks, heterogeneous clusters have become one of the trends in high performance computing. This paper focuses on the loop self-scheduling for heterogeneous clusters. Since it is possible that the cluster system is shared among many processes, the high clock rate of CPU can’t guarantee that it will provide high CPU power for the shared processes. As previous loop self-scheduling algorithms do not consider the heterogeneity of CPU and memory loading at run time, in this paper, we propose an adaptive loop self-scheduling algorithms. Based on the various CPU and memory loadings of nodes, the new method adaptively adjusts the granularity of work load assigned to the requested node. We implement our method and previously known methods on our heterogeneous cluster, and the experimental results show that our method performs better than the other policies.
|
|---|
