Decommission of HPC Allocated Systems

BC Project: FY12-02
Date of Policy: 16 Apr 2013
Last Updated: 16 Dec 2021 (see Revision Log)

In order to provide the High Performance Computing Modernization Program (HPCMP) user community with consistent policies and notification concerning decommission of allocated HPC resources, all participating HPCMP centers shall adhere to the following practices for HPC system decommissions.

User Notification

Messages can be customized to fit each center's needs, but at a minimum all centers shall send notices to users at the following lengths of time before decommission:

  • 6 months
  • 3 months
  • 1 month
  • 2 weeks
  • 1 week

Delivery should occur through multiple channels to provide the widest possible coverage. The following three communication methods reported in BC policy FY06-10 (Standardized User Notices) should be accomplished:

  • Email
  • Message Of The Day (MOTD)
  • Web Posting

Queues Access

The batch scheduler job queues will be disabled at midnight (system time) on the established decommission date. All executing jobs will be allowed to run to completion beyond the decommission date. Queued jobs that have not yet begun execution at the decommission boundary will not be allowed to begin execution. The data transfer queue will remain available for one week beyond the decommission date to allow users to archive data.

Data Access Period

Users have one week beyond the decommission date to archive or remove data from the system. At the one week boundary user login access will be terminated. Given that jobs are allowed to run up to a week after the decommission date, users are encouraged to plan accordingly as no additional time for user login access to retrieve output will be available at the end of the run.

Revision Log
Date Revision
16 Dec 2021BC Team Audit
28 Jun 2018BC Team Audit
16 Jun 2016BC Team Audit
05 May 2014BC Team Audit