Let’s build a private cloud – how hard can it be?
summary
Built a small private cloud with OpenStack on seven servers (merged control+network, two compute nodes, ~265 CPUs and 1 TB RAM), deployed with Kolla Ansible and monitored with Prometheus and Graylog. Encountered HAProxy bottlenecks (single API call spawning 500–2,000 internal queries, UI timeouts ~2.5 minutes), NUMA-induced VM slowness (fixed with libvirt numa aware or CPU pinning tradeoffs), and Ceph OSD sizing woes (resizing to ~1 TB OSDs and core‑affinitizing OSDs raises IOPS to ~50k and commit latency to ~0.04 ms). Overall: works after three‑plus months of full‑time effort and lots of monitoring.
AI-written summary · about
chapters (10)
- 0:00Introduction and speaker intro
- 1:45What is a private cloud
- 5:03Commercial vs open-source options
- 8:38Hardware and OpenStack setup
- 13:15OpenStack pros and cons
- 20:25Debugging slow API and VM performance
- 29:03NUMA and KVM performance issues
- 37:03Ceph storage performance problems
- 44:52Monitoring and troubleshooting tools
- 48:29Q&A and final thoughts
related talks
53mPrivate Cloud with OpenSource – "Now for Real!"
36mAdventures in Running a Private Cloud
17mInterview
15mThe CERN Cloud Infrastructure: 5 years in 15 minutes
33mMedia.net venturing to private cloud
38mYour Own Little HA Cloud at the Edge (or at Home)
27mPOWER your CLOUD
21mBack to the original mission: an open cloud operating system