Timely Responsive Online Scheduling Algorithm for Parallel Batch Processing

Online job scheduling has demonstrated to be a key factor impacting the output performance of batch processing systems. Existing researches either proposed solutions to a single-machine processing system, or focused on minimizing the total job completion time for multiple-machine batch processing systems. Yet the study on the approaches with the objective of minimizing the job delay time on a parallel processing system remained underinvestigated. This paper studies the online scheduling algorithms for parallel computing systems with m machines processing stochastically arrived jobs in a batch pattern. With the objective function set to minimize the maximum completion time or the total delay time of weighted jobs, respectively, we proposed two timely responsive online scheduling algorithms that yield a competitive ratio not greater than ( 5 1) / 2  . In the first proposed algorithm, with the assumption that the processing time of a batch is a constant, the jobs with higher weights get processed first so that the total completion time of weighted jobs is minimized. The objective of second proposed algorithm is to provide a timely responsive service to batched jobs by minimizing job’s total delay time. Under this algorithm we assign a priority value to each waiting job according to its weight and assume that the latency of any job does not exceed a given range. Two impact factors, i.e. the delay penalty rate and the job processing time, are projected into the algorithm to dynamically adjust job’s waiting time to minimize the value of objective function that is set as the total delay time of the processed batch. S imulation results indicate that the Minimizing Max Completion Time Algorithm yields a slightly improved performance with a competition ratio of ( 5 1) / 2  . Yet the Minimizing Total Delay Time Algorithm significantly outperforms its counterpart by projecting the two scheduling factors into the algorithm. By dynamically adjusting job’s waiting time according to its completion deadline and delay penalty factor, the objective function of minimizing total delay time is achieved and the output performance is greatly improved. KeywordsOnline Scheduling; Parallel Batch Processing; Competitive Ratio; Objective Function