GRID COMPUTING SCHEDULING BASED ON NEURAL NETWORKS

In Grid Computing environment, there are many of idle resources that compete for similar tasks. In order to gain maximum resource utilization while minimizing task completion time, a time optimization scheduling algorithm based on back-propagation neural network is proposed in this paper. The proposed algorithm predicts the submitted task run time by training neural network through a training set of samples. Each sample consist of an input vector of 7 parameters some of them to describe the resource dynamics while others for describing job characteristics and an output vector of one parameter which gives us the predicted task run time. This prediction provides reliable information for task scheduling and the resource management. The proposed algorithm is applied in compute-intensive, read- intensive and write-intensive applications. In order to evaluate the effectiveness of the proposed algorithm, the performance of the proposed algorithm is compared with Min-Min algorithm as a benchmark. Gridsim toolkit is used as a simulation environment to evaluate the performance of mentioned algorithms. Experimental results show that the proposed algorithm is better than its Min-Min counterpart in the compute-intensive, read-intensive and write-intensive applications specially when the number of jobs is increased in the experiment