1.2 Philosophy and Goals
Managers want high system utilization and the ability to deliver various qualities of service to various users and groups. They need to understand how available resources are delivered to users over time. They also need administrators to tune cycle delivery to satisfy the current site mission objectives.
Determining a scheduler's success is contingent upon establishing metrics and a means to measure them. The value of statistics is best understood if optimal statistical values are known for a given environment, including workload, resources, and policies. That is, if an administrator could determine that a site's typical workload obtained an average queue time of 3.0 hours on a particular system, that would be a useful statistic; however, if an administrator knew that through proper tuning the system could deliver an average queue time of 1.2 hours with minimal negative side effects, that would be valuable knowledge.
Moab development relies on extensive feedback from users, administrators, and managers. At its core, it is a tool designed to manage resources and provide meaningful information about what is actually happening on the system.
1.2.1 Management Goals
A manager must ensure that a cluster fulfills the purpose for which it was purchased, so a manager must deliver cycles to those projects that are most critical to the success of the funding organizations. Management tasks to fulfill this role may include the following:
- Define cluster mission objectives and performance criteria
- Evaluate current and historical cluster performance
- Instantly graph delivered service
1.2.2 Administration Goals
An administrator must ensure that a cluster is effectively functioning within the bounds of the established mission goals. Administrators translate goals into cluster policies, identify and correct cluster failures, and train users in best practices. Given these objectives, an administrator may be tasked with each of the following:
- Maximize utilization and cluster responsiveness
- Tune fairness policies and workload distribution
- Automate time-consuming tasks
- Troubleshoot job and resource failures
- Instruct users of available policies and in their use regarding the cluster
- Integrate new hardware and cluster services into the batch system
1.2.3 End-user Goals
End-users are responsible for learning about the resources available, the requirements of their workload, and the policies to which they are subject. Using this understanding and the available tools, they find ways to obtain the best possible responsiveness for their own jobs. A typical end-user may have the following tasks:
- Manage current workload
- Identify available resources
- Minimize workload response time
- Track historical usage
- Identify effectiveness of prior submissions