CRAY XE6 Batch System Layout and Limits

There are different types of queues configured on this System. These will be used to set proper priorities for different jobs, consider user permissions and resource reservations for different user groups. The configuration is laid out such that usually all you need to do is to request the number of processes (mppwidth) you need (along with the number of processes per node) and the time (walltime) for your job. The scheduling system will then place the job in the appropriate queue. Users should always specify a realistic value for the walltime. Jobs with a shorter walltime get a higher priority and may be used for backfilling (users usually specify 24h, which is the max. time limit on HLRS systems. If your job usually runs in 4h 17min and you specified 5h, your job will be selected if nodes are available for this timeframe while the job-scheduler is collecting more nodes for a larger job).

At the moment there are only two different type of queues defined:

• mpp, the default job queue for Massive Parallel Processing jobs
• ccm, the job queue for the Cluster Compatibility Mode (CCM)

If you don't specify any queue on your qsub command, then your job will be routed in the mpp queue(s).

Using the qsub option "-q ccm" your job will be prepared for the Cluster Compatibility Mode (CCM). The CCM is a software solution that provides the services needed to run most cluster-based independent software vendor (ISV) applications out-of-the-box with some configuration adjustments.

Note:

• CCM is only available for some users, not by default!
• On workdays (Mo.- Fr.) between 6:00 - 22:00, 15% of the compute nodes are reserved for short jobs with walltime lower than 4h.
• The max. walltime for jobs is 24h.

Job Run Limitations

• The maximum time limit for a Job is 24 hours.
• User limits:
  • limited number of jobs of one user that can run at the same time
  • in total a user can only allocate 60000 cores or 1875 nodes.
• User Group limits:
  • limited number of jobs of users in the same group that can run at the same time
• For CCM jobs the max. mppwidth is 256
• Batch Queue limits of all user jobs:
  • single node job queue (single): max. 300 nodes in total
  • multi node job queue (multi): max. 3400 nodes in total
  • queue for ccm jobs (ccm_base): max. 1000 nodes in total

Queues with extended wall time limits

are not available for general use. This Queue is available for Jobs, which can not run within the 24h timeframe. Access to this queue is only granted by passing an evaluation process. Following rules apply to this queue:

• Jobs may be killed for operation reasons at any time.
• Jobs will be accounted in any case. This is also true if the jab has to be terminated for operational reasons.
• Joblimit per Group = 3
• Joblimit per user = 2
• Total number of nodes/cores used for this queue = 256 / 8192
• Node types available: 32GB
• Low scheduling priority
• Max walltime 96h
• Jobs which are possible to run in a normal queue (by walltime or if the job can be splitted into subjobs have to be processed in a standard queue.

Again this queue is not for convenience but for running jobs which can not produce a result in other ways!

Frontend node for Filetransfer and unlimited runtime

a special frontend node has been set up without a limitation of the CPU time. This is necessary for jobs like:

• workflow management, an interactive process which takes care of a bunch of batch jobs. E.G. optimization tasks
• transfer of large amounts of data. Currently such jobs can not be handled on the login nodes.

By using this node, the user accepts the following rules:

• This node may be rebooted at any time
• This node may rebooted without notice
• functionality of this node may be replaced by other solutions (like 3rd party transfer, private cloud server per user/group...)
• usage of this node is at your own risk. Jobs will be charged if this node fails...
• everything which could be done on other nodes is not allowed. (e.g. compiling, data analysis, ...)
• memory usage is limited and has to be as less as possible