Optimizing grid resource allocation by combining fuzzy clustering with application preference

Abstract

Focusing on the problem of resource allocation under large-scale, distributed, autonomous, heterogeneous and dynamic environments in grid computing, a heuristic algorithm combining fuzzy clustering with application preference is proposed. Fuzzy clustering method is applied according to a group of features, which describe the user's application preference, to realize reasonable pre-classification resource. Then a resource is chosen according to the synthetic evaluation value, which can make the user's target utility maximized. There is no need to search every resource at each scheduling step. Therefore, the cost on choosing the resource to execute the current task is reduced significantly. Experimental results show that the bigger the target system, the more efficient the algorithm is, and the more satisfactorily the application preferences of users are met. Furthermore, since resources are classified by different application preferences, this method can also avoid heavy loads concentrating on only a few resources so as to improve load balance in grid environments.