How Obsolete is your Cloud?
summary
Five to six year old OpenStack clouds built on E5-era servers show clear bottlenecks: monitoring and upgrades stall on HDD boot and journaling, Ceph’s move from Filestore to Bluestore changes disk needs, and rebuilds on HDDs can take >4 days vs ~40–60 minutes on SSDs. Modern nodes use 25GbE, NVMe throughput is ~5.5 GB/s vs SATA ~500 MB/s, CPUs and memory density cut required compute nodes by roughly half. Recommend SSD control plane, NVMe-backed storage, and a small staging cloud with failure domains before migrating.
AI-written summary · about
chapters (9)
- 0:00Intro
- 1:38Overview
- 3:28Workloads
- 5:04Things have changed
- 8:48Two options
- 11:11What has changed
- 13:45Modern configuration
- 17:35What do we do
- 19:12Conclusion
related talks
29mCeph, Xen, and CloudStack: Semper Melior
24mThe Butterfly Effect of a Broken Disk: Top-Down Ceph Troubleshooting To Upstream Contribution
38mHow OpenStack became boring (and successful)
22mMoving a Large Workload from a Public Cloud to an OpenStack Private Cloud...
14mKeynote: Building and Sustaining Momentum in Open-Source Storage
19m10 years of OpenStack at CERN (From 0 to 300k cores)
46mTurning OpenStack Swift into a High‑Load Storage
29mCeph's journey at SUSE