,

Overview

The CCS HPC resources are distributed among 3 logical partitions, designated as General, Project and Dedicated Partitions to maximize throughput. Approximately 50% of the total CCS resources are available in the general partition. This partition provides a basic HPC capability to all university users on a First Come First Served basis. The remaining 50% of the CCS resources are allocated to the project partition. This partition is primarily used for large parallel jobs. Approval of an allocations committee is required to use this partition. This committee will meet a few times a year to review proposals requesting project allocations. Approved users will be allocated CPU-hours and an account number that will be used for accounting purposes. A dedicated allocation also known as the "Condo" Model is available for users requiring on demand access to resources. In this model users contribute money towards compute and/or storage. These resources are operated by CCS as part of the general pool but made available to owners on demand.

  • On demand, no wait access is available only via the "condo" model.

  • In both general and project partitions jobs are submitted and executed by a batch system which entails some wait time.

  • The General Partition is open to all users on a First Come First Served basis. Several Job queues are available (see below) for job submissions. There is a 2 simultaneous jobs/user limit on the general partition.

  • IBM p-series machines entirely allocated to project partition and all jobs need approval of the allocation committee. Approved users can submit jobs, using the account number, to the project queues until they exhaust allocated time. Jobs can be submitted to any of the Job queues in this partition and will be run as resources become available.

  • The scheduler will despatch jobs based on the requested wall clock time and job priority values in the tables below.

  • These queues/classes will be available as a part of the October, 2008 release and are prefixed according to the partition they belong to. Additional queues will be set up as necessary.

Table 10: Summary of CCS IBM p-series Job Queues
Partition Queue/Class Max CPU's Job Time Limit Job Priority
Project psmall 96 8 hrs 100
Project pmedium 256 4 hrs 90
Project plarge 576 2 hrs 75
Project pdebug 16 0.5 hrs interactive
Project pbigmem 16 24 hrs 100
Project pdata 16 0.5 hrs 100
Dedicated downer cpu's purchased unlimited immediate availability


Table 11: Summary of CCS Linux Job Queues
Partition Queue/Class Max CPU's Job Time Limit Job Priority
General gsmall 32 96 hrs 100
General gmedium 64 48 hrs 90
General glarge 128 24 hrs 75
General gdebug 8 0.5 hrs interactive
General gbigmem 8 48 hrs 100
General gdata 2 24 hrs 100
Project psmall 32 96 hrs 100
Project pmedium 64 48 hrs 90
Project plarge 128 24 hrs 75
Project pdebug 8 0.5 hrs interactive
Project pbigmem 8 48 hrs 100
Project pdata 2 24 hrs 100
Dedicated downer cpu's purchased unlimited immediate availability