Replica Prediction Guide Task Scheduling for a Highly Frequent Task-Submitting Circumstance in Grid Environment

This paper aims at studying the task scheduling in a highly frequent task submitting circumstance in a specific grid environment. The grid is mainly comprised by PCs (personal computers) which in practice will not provide replica information to the grid replica catalogue because the frequent updating operation caused by task execution will consume lots of bandwidth and computing resources. Furthermore file replacement will occur if the storage element has insufficient storage space. Considering with these problems, the Replica Prediction Guide (RPG) task scheduling algorithm is introduced. By maintaining a Replica Prediction List (RPL), a Task Queue Wait Time Queue (TQWTQ) and the total size of working directory for each storage element without holding replica information, the grid scheduler can track the data sharing pattern between a task and replica content of a host, also the task wait time can be predicted by checking TQWTQ. More practically, the RPL can predict the replica contents of a machine even different file replacement strategies actually used on it. The simulation results show that the RPG algorithm has better performance than Queue Access Cost used in OptorSim and Storage Affinity algorithm when a set of tasks are submitted in high frequency.