A novel load metric with enhanced ability of distinguishing different load status

Abstract

This paper has proposed a new metric for characterizing the load of a uniprocessor computer. This metric is defined as a ratio of the average waiting time experienced by computing tasks to the average idle time of the processor. This definition is based on two important observations. As the load becomes heavier, the average idle time of a processor becomes shorter, and the average waiting time experienced by computing tasks becomes longer. The new load metric aims to have an enhanced ability in distinguishing different load status. The valid comparison of different load status using the new load metric is subject to the satisfaction of certain conditions: a majorization has to be satisfied between two processes of the measures (idle time or waiting time). The ability of the new load metric on distinguishing different load status has been validated through numerical simulations. The results showed that the new load metric can facilitate balancing the load among multiple servers better than the conventional load metric of average utilization, when they are used as the task assignment criteria.