Overview

Once the Great Lakes cluster is ready, it will be configured and provided as a preview for early users. This date is tentatively set for July; see the Great Lakes timeline for updates and other milestones.  If you are interested in becoming an early user on Great Lakes, please let us know at hpc-support@umich.edu.  Some points to consider:

  • Great Lakes uses the Slurm scheduler, while Flux uses Torque and Moab.  You will have to rewrite your job submission scripts.
  • The initial maximum walltime for jobs will be 48 hours, which we will increase later in the testing period.
  • Great Lakes has different /home and /scratch directories than Flux.  You will need to copy any data from Flux to Great Lakes. As a recommendation, use Globus to do these copies.
  • You may experience bugs or inconsistencies during this period.  We want feedback on your experience and detailed tickets if you encounter any issues.
  • There may be service outages while we finalize the configuration, including Slurm configurations and settings. We will give every attempt at announcing any required outages to you.
  • Not all software has been built for Great Lakes.  If software you need is not available, please submit a ticket to hpc-support@umich.edu.
  • The HDR InfiniBand speed is 100 Gb/s speed.  As a comparison, InfiniBand speeds on Flux range from 40-100 Gb/s.
  • During this preliminary phase, there will be no charge for using Great Lakes.

Who Should Consider being an Early User

There are a few reasons why you might want to consider being an Early User on Great Lakes:

  • You have had performance issues previously on Flux
  • You have large parallel codes that you want to verify work with the new InfiniBand network
  • You have code that you want to verify works on the new Intel Skylake processors
  • You want to get a jumpstart on migrating your workflows to Slurm and Great Lakes

Using the Cluster

For a step-by-step explanation on how to get started and a reference cheat sheet, see the Great Lakes User Guide.  If you want a more in-depth guide to Slurm commands and features, see the Slurm User Guide.

Feedback

If you encounter an issue or have any kind of feedback (positive or negative), please email hpc-support@umich.edu detailing your situation.