HPC BoF Discussion
summary
Discussion lists concrete priorities: adopt Spack-based builds to produce CPU-optimized math libraries rather than generic RPMs; run benchmarks with those builds; expand MPI testing after finding OpenMPI outperforms MPICH in current Arm tests and evaluate RDMA/InfiniBand use and switch donation needs; deploy OpenHPC via Ansible and run CI in a multi-vendor lab to catch regressions; validate distributed storage with Lustre client testing; pursue QEMU SVE work with QEMU SVE user-space emulation for early software validation.
AI-written summary · about
chapters (20)
- 0:00Introduction and agenda
- 1:31Spack vs RPM integration
- 3:20MPI and RDMA testing
- 4:28Ansible for OpenHPC deployment
- 7:03Networking and VLAN setup
- 9:23Lustre file system plans
- 11:13HPC objectives and CI goals
- 13:25Spack advantages for libraries
- 16:39Lustre performance questions
- 17:44Networking in HPC use cases
- 19:31MPI and networking priorities
- 21:18SVE QEMU emulation update
- 24:15SVE roadmap and support
- 26:45SVE emulation models explained
- 31:35Missing packages and differences
- 33:50Additional test workloads
- 35:37Open source HPC collaboration
- 37:57Performance benchmarking policy
- 40:54Prioritizing vendor-neutral changes
- 41:37Q&A and wrap-up
related talks
46mBuilding Supercomputers Using Kubernetes and OpenHPC
20mBenchmarking HPC workloads on OpenShift
23mImplementation Challenges: From HPC to Containers in the Academy
08mLightning Talk: Orchestrating Kubernetes Clusters on HPC Infrastructure
29mTowards Sustainable Multi-Cluster Management in Real-World Academic Infrastructure
24mPropagating Programming Paradigms: Lifting High Performance Compute Into K8s
10mOvercome Interconnect Bottlenecks in HPC Clusters
01h 21mHPC in the cloud