Moab Workload Manager

1.3 Workload

Moab can manage a broad spectrum of compute workload types, and it can optimize all four workload types within the same cluster simultaneously, delivering on the objectives most important to each workload type. The workload types include the following:

1.3.1 Batch Workload

Batch workload is characterized by a job command file that typically describes all critical aspects of the needed compute resources and execution envionment. With a batch job, the job is submitted to a job queue, and is run somewhere on the cluster as resources become available. In most cases, the submittor will submit multiple batch jobs with no execution time constraints and will process the job results as they become available.

Moab can enforce rich policies defining how, when, and where batch jobs run to deliver compute resources to the most important workload and provide general SLA guarantees while maximizing system utilization and minimizing average response time.

1.3.2 Interactive Workload

Interactive workload differs from batch in that requestors are interested in immediate response and are generally waiting for the interactive request to be executed before going on to other activities. In many cases, interactive submittors will continue to be attached to the interactive job, routing keystrokes and other input into the job and seeing both output and error information in real-time. While interactive workload may be submitted within a job file, commonly, it is routed into the cluster via a web or other graphical terminal and the end-user may never even be aware of the underlying use of the batch system.

For managing interactive jobs, the focus is usually on setting aside resources to guarantee immediate execution or at least a minimal wait time for interactive jobs. Targeted service levels require management when mixing batch and interactive jobs. Interactive and other jobs types can be dynamically steered in terms of what they are executing as well as in terms of the quantity of resources required by the application. Moab can apply dynamic or malleable job facilities to dynamically grow and shrink jobs as needed to meet these changing constraints.

1.3.3 Calendar Workload

Calendar workload must be executed at a particular time and possibly in a regular periodic manner. For such jobs, time constraints range from flexible to rigid. For example, some calendar jobs may need to complete by a certain time, while others must run exactly at a given time each day or each week.

Moab can schedule the future and can thus guarantee resource availability at needed times to allow calendar jobs to run as required. Furthermore, Moab provisioning features can locate or temporarily create the needed compute environment to properly execute the target applications.

1.3.4 Service Workload

Moab can schedule and manage both individual applications and long-running or persistent services. Service workload processes externally-generated transaction requests while Moab provides the distributed service with needed resources to meet target backlog or response targets to the service. Examples of service workload include parallel databases, web farms, and visualization services. Moab can apply cluster, grid, or dynamically-generated on-demand resources to the service.

When handling service workload, Moab observes the application in a highly abstract manner. Using the JOBCFG parameter, aspects of the service jobs can be discovered or configured with attributes describing them as resource consumers possessing response time, backlog, state metrics and associated QoS targets. In addition, each application can specify the type of compute resource required (OS, arch, memory, disk, network adapter, data store, and so forth) as well as the support environment (network, storage, external services, and so forth).

If the QoS response time/backlog targets of the application are not being satisfied by the current resource allocation, Moab evaluates the needs of this application against all other site mission objectives and workload needs and determines what it must do to locate or create (that is, provision, customize, secure) the needed resources. With the application resource requirement specification, a site may also indicate proximity/locality constraints, partition policies, ramp-up/ramp-down rules, and so forth.

Once Moab identifies and creates appropriate resources, it hands these resources to the application via a site customized URL. This URL can be responsible for whatever application-specific hand-shaking must be done to launch and initialize the needed components of the distributed application upon the new resources. Moab engages in the hand-off by providing needed context and resource information and by launching the URL at the appropriate time.

See Also