Study of a machine selection algorithm for grid computing

For the heterogeneity and distribution of grid computational resources in many aspects, it is important to select a resource subset from all available resources to support an application execution. If assigning the suitable resources to an application, the efficiency of the application will be improved. Usually, a complicated application includes many inherences, and such an application is suitable for execution in machines with different architecture. Based the monitoring results of the resources, a machine selection algorithm with different performance criteria is designed using fuzzy clustering methods. This algorithm clusters all available nodes into several logical groups, and each group is called a logic cluster here. According to the difference of application types, one or several logical clusters suitable for them, with high clustering center value, are selected using the λ-matrix method. The objective of the machine selection based on the application types is to improve the performance of the application scheduler. Test shows that the computation-intensive application is suitable for execution in the logical cluster with high computing capability, and the communication-intensive application is suitable for the logical cluster with high communication performance.