Keynote: OpenStack and the Software-Defined Supercomputer
summary
Argues that OpenStack becomes a software-defined supercomputer by combining bare metal and virtualized nodes into one reconfigurable cloud; implements this with Nova and ironic, Neutron/ovn, Magnum, Blazar, Keystone, Manila, CloudKitty and an Azoth portal. Uses CPU pinning, hugepages, NUMA-aware placement, SR-IOV with ovn offloads and GPUs to hit ~49 Gbps and ~1.6 µs local latency, 180 Gbps aggregate in tests, and up to ~200 Gbps memory bandwidth per VM; deploys Jupyter/Dask, Kubernetes and Slurm platforms and reports functional production use.
AI-written summary · about
chapters (8)
- 0:00Introduction and background
- 3:06OpenStack for research computing
- 7:41Software-defined supercomputer demo
- 11:54Virtualization performance tradeoffs
- 16:48High-performance networking
- 18:57User-friendly research platforms
- 25:02Batch compute environments
- 26:06Community and closing
related talks
49mIntroduction to OpenStack
56mKeynote on Building the Cloud for You
43mSupporting NASA Science with OpenStack
21mKeynote: CERN Experiences - Ricardo Rocha & Clenimar Filemon
18mOpenStack Playing for an Open World
14mKeynote: Science at Light Speed: Cloud Native Infrastructure for Astronomy Workloads
30mKeynote: Cloud Native Networking
01h 21mModern Infrastructure: The Convergence of Network, Compute, and Data