HFSS15: HPC Job Properties

A number of properties that apply to the job as a whole may be specified when the job is submitted. Some of the main job properties are:

• Job Name - text string that may be used to identify or describe the job

• Job Template - Template that controls job property defaults and allowed values. (See Windows® HPC Job Templates)

• Priority - one of five levels: Lowest, BelowNormal, Normal, AboveNormal, and Highest

• Job Unit Type - computing resource unit granularity: core, node, or socket (See Selecting Computation Resource Units (Job Unit Type).

• Minimum and maximum number of computing resource units for the job. (See Specifying the Number of Compute Resource Units for Windows® HPC Jobs).

• Exclusive - if true, no other jobs will be scheduled on nodes allocated to this job

• Node group names - run the job on nodes belonging to these node groups

• Node list - if specified, runs the job only on the nodes in this list

Set the minimum/maximum number of computing resource units to the number of units the jobs needs. For example, set the number of units to sixteen, to be able to run four distributed engines, each requiring four cores for multi-processing.

The Priority property controls which jobs are run by the scheduler when several jobs are ready to run. The Exclusive property allows the user additional control on the resources used by a job. For example, a job may be scheduled which requires all memory on a host, but only some of the cores on the host; the Exclusive property would prevent other jobs from being scheduled on the unused cores on the host. Node groups are created by cluster administrator to group together a set of similar nodes. For example, a node group '32GB Nodes' could be created to consist of all nodes, which have 32GB of available memory. Jobs that are memory intensive are run on nodes that belong to this node group.

There are other job properties that further limit which nodes may be allocated to a job:

• Minimum and/or maximum number of cores per node

• Minimum and/or maximum memory per node: generally expressed in megabytes (MB)

The minimum limits may be used to ensure that the nodes allocated to this job have sufficient resources (nodes or memory) to efficiently run the job. The maximum limits may be used to optimize use of the cluster resources. For example, a maximum memory limit may be used to ensure that nodes having a large amount of memory are only used for jobs that actually need a large amount of memory and are not allocated to jobs that only require a limited amount of memory. Setting these limits acts as a filter on the nodes which may be allocated to a job; they do not ensure that the resources will be available to the current job. For example, setting a minimum memory per node of 2000 MB ensures that nodes allocated to the current job have at least 2000 MB of memory. A node with 2000 MB of memory may be allocated to the current job, even if another job running on the node is using 1500 MB of memory, however.

There are several other job properties that are not described here. See the Microsoft documentation for more details.