Skip to Main Content

The University of Tennessee

Newton header banner

Frequently Used Tools:



Home » Documentation » Using the Grid Engine » Job Throttling

Job throttling

The term "job throttling" means to place limits on the number of CPUs that a user or user group can execute jobs on at any one time. The throttling policy ensures that the cluster is not monopolized, and it allows us to guarantee that a given job will execute in a reasonable amount of time given a sufficiently high priority. To this end, we use job runtime information, group membership, and specific user settings to place limits on the number of CPUs that can be used. Here is a hypothetical example:

| Group | User | Run Time | Limit | | Physics | * | short | 100 | | Physics | * | medium | 50 | | Physics | * | long | 25 | | English | * | short | 200 | | English | * | medium | 100 | | English | * | long | 50 | | English | Bruce | * | 12 | | Users | * | long | 0 |

This means that your choice of job runtime will determine how many CPUs you are able to use. For this reason, it is important to know what your group limits are. These can be determined as follows:

[username@newton ~]$ qconf -srqs long
{
   name         long
   description  NONE
   enabled      TRUE
   limit        projects English queues long* to slots=50
   limit        projects Physics queues long* to slots=25
   limit        users {@users} queues long* to slots=0
}

You can similarly view the current limits for "medium" and "short" jobs.

Back to Using The Grid Engine