Illinois Campus Cluster Program

Illinois Campus Cluster Program

Keeping data secure

The Beckman Institute is an Investor with the Illinois Campus Cluster Program. This investment provides Beckman researchers with access to computing resources through the Beckman primary queue described below. If you are a researcher affiliated with the Beckman Institute, you may request access via the ICCP New User Form and select “Beckman” from the dropdown menu for “Primary Queue.”

Data storage UI-03-160809-034

Beckman primary queue resources

The Beckman Primary Queue is supported by the following hardware:

  • Nine, Intel chip-set based compute nodes
  • Four 256G Haswell nodes with 24 cores per node
  • Five 256G Broadwell nodes with 28 cores per node
  • Note: Jobs compiled to run on either chip type should run unmodified on the other chip set. Specialized code or library use could change that, of course
  • There are no GPUs or accelerators in any of the current nodes
  • All the nodes have an IB network interface 12tb of shared disk storage.

Queuing/batching policies

  • ICCP implements the "FairShare" scheduling policies. In a nutshell, FairShare prefers small jobs and fewer active users. There are in-depth discussions of FairShare elsewhere.
  • ICCP offers investors 100% use of the resources they purchase. It is rare that 100% use is attained. The "unused" portions of all investor resources are made available to all ICCP users in the form of a "Secondary Queue." The secondary queue strictly limits the run times of jobs to two hours or less. However, the available nodes can contain GPU or other special accelerators.
  • The Beckman Primary Queue has additional restrictions designed to enable higher throughput of short jobs:
    • Jobs that request more than 336 hours of run time (more than two weeks) will only be queued to the Broadwell nodes
    • Jobs that request less than 336 hours should run on any available node.
  • NOTE: ICCP has a scheduled maintenance period each month. The queuing system will not accept jobs that could run into the maintenance period.

Before you join

The Beckman Primanry Queue is a shared resource. It would be helpful if you have an idea of:

  • The disk space you will consume
  • The node/core counts you expect to use in a typical job
  • How long you anticipate your jobs will run
  • Have a plan for managing the data you add/create once your experiments are complete

If you find that your initial estimates are way off, that is OK. We ask that you let us know as you learn about what works for you. If you realize that you need a lot more resources in storage or computing power, please let us know before you consume them because otherwise you would most likely be clobbering other Beckman citizens.

We need lead time to arrange for additional resources and your help in figuring out how much more we need.

Other Illinois high performance computing resources

If you have questions, email the Beckman Help Desk.