OpenPOWER and GPU Porting Platform for FOSS Using OpenStack at OSUOSL
summary
OSUOSL runs an OpenStack cluster (Queens) on POWER8 and POWER9 hardware serving about 100 FOSS projects, hosting ~265 VMs across nodes and tracking 45 VMs per hypervisor. Migrating VM disk to Ceph mimic 300 TB cuts I/O contention and enables live migrations. A homegrown CI stack uses Jenkins + NVIDIA Docker with four CI VMs and two P100 GPUs (loaned) for tensor workloads, integrated via GitHub OAuth; upgrades and kernel tuning remain active operational work.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 0:53What is the OSUOSL
- 2:22What do we provide
- 3:13History with OpenPOWER
- 5:00What do FOSS projects need
- 8:06Projects using OpenPOWER
- 8:43Ticketing system
- 9:38Technical side
- 9:53OpenStack
- 12:10Hardware
- 13:04Storage
- 15:12Hypervisor
- 16:05Dashboard
- 17:05Overview
- 18:05Cluster Dashboard
- 18:29Initial Challenges
- 20:55Other Challenges
- 23:21Continuous Integration
- 24:40VM Usage
- 25:30CI on GPUs
- 27:15GPU usage graph
- 27:44Upcoming changes
- 30:13Questions
related talks
43mSupporting NASA Science with OpenStack
16mBOSH & GPU - a Match Made in Heaven
06mPOWER your Cloud
15mAn open cloud for european researchers, staff and students
01h 01mDoes Green Software need Open Hardware?
29mTowards Sustainable Multi-Cluster Management in Real-World Academic Infrastructure
16mA novel Cloud High Performance Computing infrastructure for user
27mKeynote: OpenStack and the Software-Defined Supercomputer