1.1 Value of a Batch System
Batch systems provide centralized access to distributed resources through mechanisms for submitting, launching, and tracking jobs on a shared resource. This greatly simplifies use of the cluster's distributed resources, allowing users a single system image in terms of managing jobs and aggregate compute resources available. Batch systems should do much more than just provide a global view of the cluster, though. Using compute resources in a fair and effective manner is complex, so a scheduler is necessary to determine when, where, and how to run jobs to optimize the cluster. Scheduling decisions can be categorized as follows:
1.1.1 Traffic Control
A scheduler must prevent jobs from interfering. If jobs contend for resources, cluster performance decreases, job execution is delayed, and jobs may fail. Thus, the scheduler tracks resources and dedicates requested resources to a particular job, which prevents use of such resources by other jobs.
1.1.2 Mission Policies
Clusters and other HPC platforms typically have specific purposes; to fulfill these purposes, or mission goals, there are usually rules about system use pertaining to who or what is allowed to use the system. To be effective, a scheduler must provide a suite of policies allowing a site to map site mission policies into scheduling behavior.
1.1.3 Optimizations
The compute power of a cluster is a limited resource; over time, demand inevitably exceeds supply. Intelligent scheduling decisions facilitate higher job volume and faster job completion. Though subject to the constraints of the traffic control and mission policies, the scheduler must use whatever freedom is available to maximize cluster performance.