The goal of workload management software is to make certain the submitted jobs ultimately run to completion by utilizing cluster resources according to a supplied policy. But in order to achieve this goal, workload management systems usually must perform some or all of the following activities:
The typical relationship between users, resources, and these workload management activities is depicted in Figure 14.1. As shown in this figure, workload management software sits between the cluster users and the cluster resources. First, users submit jobs to a queue in order to specify the work to be performed. (Once a job has been submitted, the user can request status information about that job at any time.) The jobs then wait in the queue until they are scheduled to start on the cluster. The specifics of the scheduling process are defined by the policy rules. At this point, resource management mechanisms handle the details of properly launching the job and perhaps cleaning up any mess left behind after the job either completes or is aborted. While all this is going on, the workload management system is monitoring the status of system resources and accounting for which users are using what resources.