Recommendations for migration to new systems

From CC Doc
Jump to: navigation, search
Other languages:
English • ‎français

Parent page: Migration to National Systems

Legacy system operations

Compute Canada will replace most of its legacy systems with new national systems that will consolidate resources and centralize services. Some of the legacy systems may continue in operation at the discretion of the institution or region which manages each system. Please ask the relevant institution or region if you want to know the plans for a specific system. A list of regional contacts can be found here.

2017/2018 and 2018/2019 mix of systems

The 2017/2018 and 2018/2019 resource allocation years are transition years with a mix of legacy and new systems available for use. Users on systems scheduled for decommissioning will have to choose which of the new or remaining legacy systems to move to.

  • New systems have better performance than old systems, but how much better depends strongly on the application. See "Specific software" below for available data.
  • Some legacy systems, though not yet decommissioned, will operate with limited vendor support and may suffer decreased reliability. Capacity (that is, the number of cores) is expected to decrease gradually as nodes fail and cannot be repaired. Contact regional support for the outlook for specific systems.
  • Cloud resources are available for users with very specific and customized software requirements, and for platforms and portals.

Cedar and Graham are very similar, general purpose, large systems. There is no particular difference which would recommend one over the other.

  • Hardware: There are different proportions of small- and large-memory nodes, and nodes with GPUs.
    • The high-performance interconnect is Intel OmniPath at Cedar and Infiniband at Graham. Performance is expected to be very similar.
    • For more details see the Cedar and Graham pages.
  • Software will be made available through a universal, standardized process. There will be minimal differences between the software available on each machine.

General migration recommendations

  • If you have a RAC allocation then your allocation letter indicated which systems you were allocated. We recommend that you move to your allocated system or systems as soon as possible.
    • Compute Canada is open to requests for migration to other machines if the allocated system is unsuitable in some way.
  • If you do not have a RAC allocation and your current system is being defunded/deactivated, then we recommend that you move to one of the new systems Cedar (GP2) or Graham (GP3) as soon as they become available.
    • Every user will be able to acquire a Rapid Access Service (RAS) account on any national system.
    • As a rule-of-thumb those users in western Canada (Saskatchewan to BC) should choose Cedar (GP2), and the rest should choose Graham (GP3). This has no particular technical justification; we merely hope this guidance will roughly balance the demand for these two systems.
    • You may also move to a different legacy system. This will be easiest if you are already familiar the legacy system. Please contact your regional support for details.
    • We encourage you to use the Rapid Access Service (RAS) to test your application's performance on new systems as early as possible.

You are encouraged to send trouble reports and requests for adjustments to as soon as you detect a problem. Compute Canada will provide in-depth consultation and support for migration issues.

RAC 2017

Mar.30/2017: RAC award notifications were sent out.

  • In order to balance the load across the mix of available systems the Resource Allocation Committee found that considerable re-arrangement and re-allocation of your requests was needed.
  • Compute Canada will monitor performance and load on all machines, and anticipates that re-allocations may be necessary during the 2017/2018 RAC year.
    • Compute Canada will notify you if we believe a re-allocation is desirable.
    • Compute Canada will be open to requests for migration to other machines if the allocated system is unsuitable in some way.
  • The number of core-years allocated is based on the performance of legacy systems. New systems will be faster, but no adjustments were made because there has been no opportunity to carry out the measurements needed to make rational adjustments.

RAC 2018

Sep.12/2017: The RAC 2018 application portal is expected to open on Oct.3. Please see the CC Research Allocation Competitions portal for further details.