Skip to main content

SECOND COMMUNICATION – Transition to new role for the Perceval Supercomputer

Les Michelson
Thursday, January 16, 2020 at 3:31 pm

Please read the attached document which details the changes in the support model and roles for the Perceval Supercomputer to be effective March 1, 2020. My apologies in advance for the length of this communication, but I encourage you to review it as some changes may require action on your behalf or that of your PI/mentor. Please do not hesitate to contact me should you any questions concerning the upcoming changes planned for Perceval. Thank you.

Please share this communication with the PI(s) in your lab as necessary

Dear Perceval User:

As Principal Investigator on the NIH grant that funded the Perceval supercomputer for Rutgers University I am delighted and proud to tell you that over the more than 4.5 years this instrument has been operational, a very large body of work in the life sciences has been facilitated, resulting in numerous publications and undoubtedly leveraging many subsequent and successful grant proposals. Perceval also demonstrated that a large-scale, shared computational facility is viable and capable indeed of supporting and being effectively managed for a sizable community of Rutgers investigators. The Office of Advanced Research Computing and its parent, the Office of Information Technology, deserve great credit for their effort and support which, in very large measure, contributed to this success.

On March 1, 2020, Rutgers’ contractual obligation to support Perceval will officially end. At this time Perceval will be re-purposed to support a broader scope of activities including but not limited to research, teaching and pre-production hardware and software testing for the wider research community. Perceval will also support expanded open access, including a larger non-pre-emption pool for the Amarel Distributed Linux Cluster. OARC will provide more detailed information regarding these changes and opportunities as the March 1 date approaches.

We, of course, want this transition to have as little impact on the continuity of your research as possible. OARC manages several major resources that can assist in achieving this result. Among them are Amarel ownership and open access. Amarel ownership provides discrete resources dedicated to your lab and immediately accessible whenever you need it. Owners purchase nodes and storage with all other infrastructure provided and technical and scientific support delivered and at no cost by OARC. Amarel Open Access is another option and has no associated cost. There are several restrictions regarding Amarel Open Access, nevertheless it is an excellent solution for many research computing requirements. OARC plans to have other advanced computational resources in production over the next year and will communicate them to you as they become available.

Investigators who collaborated on the NIH proposal (major and minor users) that funded Perceval are considered owners for the purpose of retaining all or part of the project storage allocated to them, their students and lab staff by purchasing Amarel project storage up to their allocated capacity as of March 1, 2020. Currently, Amarel storage cost is $150/TB for a four-year term. These investigators will retain their allocations until 10/31/20. Those wishing to purchase storage must do so by 10/31/20 or arrange to move their storage out of the Amarel environment. The four-year storage term will begin 11/1/20 for these purchases. Home directory storage and scratch storage are not impacted by the Perceval transition and no action is required by the account holder to retain those allocations. Please contact Kristin Lepping (klepping@oarc.rutgers.edu) for further information if you want to retain all or part of your storage allocation after 10/31/20. A separate notice will be communicated to the Perceval proposal collaborators to alert them to the transition and review available options.

Please note that as of 3/1/20, Perceval accounts will be treated as Open Access subject to pre-emption; i.e., jobs submitted by these accounts may be terminated if insufficient resources are available to satisfy owner entitlement for their own jobs. The non-pre-emptible partition is an available option if your jobs are unable to tolerate pre-emption. OARC plans to expand the non-pre-emptible partition utilizing a portion of the Perceval resources.

    In summary:

  • Perceval transitions from an instrument supporting only NIH-funded work to a more general role;
  • Investigators who were collaborators on the proposal that funded Perceval are eligible to retain all or part of the storage allocated to them as of 02/28/20. This storage must be purchased by 10/31/20;
  • Home directory and scratch storage are not impacted- only project storage;
  • Jobs submitted to Perceval by non-Amarel owners are subject to pre-emption. Use the non-pre-emptible partition to avoid re-running jobs or if automatic re-queuing upon pre-emption is not a suitable option for your applications. Although the non-pre-emptible partition will be expanded, users should expect longer queue times versus the pre-emptible main partition. Please contact us if you have questions regarding pre-emption. We will be happy to work with you to find the best approach.

 The shelf life of technology, as we all know, is fleeting. Computational technology is perhaps among the more extreme examples of this. On the other hand, we continue to look for new opportunities, e.g., NIH S10, NSF MRI and similar programs to expand and enhance Rutgers cyberinfrastructure arsenal. The proposal that funded Perceval could not have succeeded without collaboration across the Rutgers research communities. We look forward to exploring new programs with you.

Thank you once more for helping to make Perceval a great success through the work you have done on it. Do not hesitate to contact me should you have any questions about this or other changes that have been detailed in this communication.

Regards,

Les


Leslie P. Michelson, Ph.D.

Leslie.Michelson@rutgers.edu
Office of Advanced Research Computing
Rutgers University