Periodic Multi-installment Load Partition, Distribution and Result Collection

摘要

Multi-installment scheduling (MIS) has been proved to be an effective strategy for minimizing finish time of large-scale workloads on network-based computing systems. However, existing studies all assumed that the amount of result could be neglected compared to that of the original workloads. In this paper, we consider a general case where a large amount of result should be retrieved back to users from parallel computing units. We propose a periodic multi-installment scheduling model to obtain an optimal load partition, distribution, and result collection strategy. Experimental results show that our scheduling strategy could greatly reduce the finish time of workloads, inclusive of processing time and result collection time.