Evaluation of relative-class based job assignment

摘要

The conventional size-interval (SI) based job assignment only depends on the job's size for classifying its execution priority. In this paper, we consider computational and deadline requirements of jobs in order to decide the relative priority (i.e., class) of each job in a batch. To evaluate asymptotic behavior of a modified SI algorithm with the relative class, we propose a new queueing model. The simulation result shows that the fundamental limiting boundary of the proposed job assignment algorithm can be achieved with respect to the mean response time in distributed and heterogeneous resources.