A Fast and Intelligent Resource Allocation Service for Service-Oriented Grid

Grid has evolved dramatically into the era of service-oriented grid, which facilitates building of large-scale systems in standard fashions, reusability of essential functions, and interoperability among components. However, grid resource allocation is still a challenging problem for which a grid scheduler has to be operating in a dynamic and uncertain environment. Conventional scheduling algorithms will fail due to the static rules used and much user intervention required. We suggest that learning with neural networks is promising to solve this problem. In this paper we propose a fast and intelligent resource selection algorithm based on neural networks. Extreme Learning Machine (ELM) is exploited as the learning paradigm due to its fast learning speed and satisfactory performance. Moreover, we present an architecture, which defines components of a proposed resource allocation service and also specifies interactions to the other service-oriented grid components. Experiments show that the proposed scheduling algorithm outperforms the conventional algorithm in terms of computing power utilization.