Presentations from LSF User Group @ SC19

On November 22nd, 2019 we held our annual LSF user group at SC19 in Denver.

IBM Distinguished Engineer Khalid Ahmed kicked off the afternoon discussing some of the challenges with running HPC workloads in Kubernetes.  He explained how LSF can be used as an enhanced pod scheduler in both a pure Kubernetes and hybrid HPC-Kubernetes environment.  The latter option enabling Kubernetes workloads to be introduces into existing HPC environments. This capability is now available in Service Pack 9.

Pavel Roy Paladhi from IBM’s semiconductor design team gave an in depth talk on the challenges associated with signal integrity.  The fastest simulation is the one you don’t need to run, and Pavel showed that using GPU accelerated IBM Bayesian Optimization (currently available as a tech preview) a solution could be reached in as little as 40 jobs, compared to the usual 5625 jobs the default approach would have taken – a 100x improvement in time to solution.

Next up was John Welch from the Washington Systems Lab. John demonstrated using IBM Watson ML Community Edition through LSF’s Application Center web portal.  He focused on complexity of running containerized AI frameworks could be abstracted to a simple point and click interface. Jupyter Notebooks and Tensorflow were used as examples.  John’s demonstrations will be available on YouTube soon.

The penultimate session was given by Eric Zhang from the IBM Federal CTO office who discussed the challenges associated with workload and energy efficiency at exascale.

Finally, I provided an overview of the new capabilities released in during 2019 in Service Packs 8 & 9 and outlined some of the areas we will be focusing on in 2020.

Thank you to all who took time out of your busy SC19 schedule to attend, and I hope we see you all again next year.   If you are interested in presenting at a future LSF user group, please get in touch.