OpenStack for Computational workloads
summary
Practical checklist of OpenStack changes that improve real HPC/ML performance: enable NUMA awareness and huge pages to cut memory latency, use CPU pinning and host passthrough or defined feature requests (AVX etc.) instead of emulation, and prefer PCI passthrough or vGPU for accelerators. Network with SR-IOV/DPDK and avoid routed storage; favor direct local IO and tune flavors to avoid memory overcommit. Validate with HPL, STREAM triad, MPI tests and iperf; a reference deployment named Thermoscript is provided.
AI-written summary · about
chapters (8)
- 0:00Introduction to OpenStack HPC
- 1:50Computational workload use cases
- 3:58High-priority performance configurations
- 6:25Networking options for HPC
- 7:51Storage considerations for HPC
- 9:19Performance monitoring and tools
- 10:06Reference configuration case study
- 11:36Q&A and closing
related talks
26mWhy the Next Generation of HPC systems Needs Open Source Driven Standardization
23mOpen Matters: HP & OpenStack
01h 50mKeynote
22mMoving a Large Workload from a Public Cloud to an OpenStack Private Cloud...
27mLessons Learned from a Large Scale OpenStack Deployment with Tripleo
46mTurning OpenStack Swift into a High‑Load Storage
30mOpenStack Works, so now what?
14mKeynote: Open Source, Better, Faster, Stronger