Pseudo-cluster based resource selection for data parallel applications in computational grid

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.