数据并行应用的基于伪机群的网格资源选择

Pseudo-cluster Based Resource Selection for Data Parallel Applications in Computational Grid

为给数据并行应用选择适合的网格资源,文章给出一个伪机群的想法。非专用机群中的主机可根据资源性能偏差率构造出若干伪机群,使同一个伪机群中的资源性能彼此相近。专用机群也可被视为伪机群,故对于任何类型的应用,伪机群思想使网格资源彼此有了可 比性。实验表明,拥有最高整体性能的伪机群将使应用取得最短的运行时间。

To select suitable grid resource for a data parallel application, an idea of pseudo-cluster is introduced. According to resource performance deviation rates, the hosts in an undedicated cluster can compose some pseudo-clusters so that resource performances are close to each other in an identical pseudo-cluster. Dedicated clusters can also be considered pseudo-clusters, thus the idea of pseudo-cluster makes grid resources comparable with each other for any type of applications. The pseudo-cluster with highest integral performance will have applications obtain shortest runtime, which is proved by experimental results.