A grid resource discovery method based on adaptive k-nearest neighbors clustering

Abstract

Several features of today's grid are based on centralized or hierarchical services. However, as the grid size increasing, some of their functions especially resource discovery should be decentralized to avoid performance bottlenecks and guarantee scalability. A novel grid resource discovery method based on adaptive k-Nearest Neighbors clustering is presented in this paper. A class is formed by a collection of nodes with some similarities in their characteristics, each class is managed by a leader and consists of members that serve as workers. Resource requests are ideally forwarded to an appropriate class leader that would then direct it to one of its workers. This method can handle resource requests by searching a small subset out of a large number of nodes by resource clustering which can improve the resource query efficiency; on the other hand, it also achieves well scalability by managing grid resources with adaptive mechanism. It is shown from a series of experiments that the method presented in this paper achieves more scalability and efficient lookup performance than other existing methods. © Springer-Verlag Berlin Heidelberg 2007.