In this paper, user-level job placement system using new load index calculation scheme is proposed. In proposed load index calculation scheme, memory load is considered as well as CPU load to calculate load index of each node of clustered systems, also, characteristics of incoming job is considered.
Load index of previous load balancing systems is CPU based. Load of each node is calculated by counting running jobs in node. Memory resource is increasingly important because of rapid improvement of CPU and many data access of application. In this situation, no consideration of memory resource in load balancing system can cause not efficient utilization of all system resource. If memory requirement of jobs in a node exceeds main memory of the node, execution time of jobs is increased by latency of paging or swapping. Also, previous load balancing systems calculate load index from machine’s point of view. Only machine’s resource usage is considered and job characteristics is not considered. So, suitable placement from job’s point of view is impossible. We propose more accurate load index calculation scheme and implement job placement system using the load index calculation scheme considering memory as well as CPU usage and job characteristics.
We evaluate performance of our proposed system in three point of view; load imbalance, slowdown, total execution time. Compared with Zhang’s CPU-MEM based[6] and CPU based, MEM based job placement system, our proposed system shows more accurate job placement and reduced job execution time.
