Software Engineering Radio - the podcast for professional software developers SE Radio 709: Bryan Cantrill on the Data Center Control Plane
5 snips
Feb 26, 2026 
Bryan Cantrill, co-founder and CTO of Oxide Computer and former Joyent CTO famed for DTrace, talks data-center control planes. He recounts hyperscale surprises, firmware and BMC pitfalls, and why public-cloud hardware diverges from commodity gear. He explains what a control plane does, why Joyent struggled with Node.js, and why Oxide chose Rust and Illumos for a vertically integrated rack design.
AI Snips
Chapters
Transcript
Episode notes
Drive Firmware Caused Massive IO Outliers
- Joyent discovered pathological IO tail latency at a Samsung-scale DC caused by substitute Toshiba drives slipped in by the OEM instead of HGST drives.
- The Toshiba drives had a firmware bug that stopped acknowledging reads for ~2.7 seconds, exposing lack of control when vendors mix parts.
Why Hyperscalers Design Their Own Hardware
- Public cloud vendors design their own machines because vertically integrated systems eliminate the brittle mix-and-match failures seen with commodity OEM stacks.
- Oxide's thesis: cloud-like infrastructure should be purchasable on-prem by delivering integrated rack-scale hardware and software.
BMCs Are A Shadow Attack Surface
- Baseboard management controllers (BMCs) are a fragile, shadow computing layer with proprietary firmware and hardcoded root credentials that create security and operational burdens.
- BMCs run separate networks and require patching/management, often hidden from customers and causing severe risks if compromised.