Software Engineering Radio - the podcast for professional software developers SE Radio 709: Bryan Cantrill on the Data Center Control Plane
Feb 26, 2026
Bryan Cantrill, co-founder and CTO of Oxide Computer and former Joyent CTO, system engineer known for DTrace and data-center work. He talks about hidden hardware variation and firmware failures, why hyperscalers build custom hardware, flaws in baseboard management controllers, what a control plane does, why Oxide chose Rust and Illumos, and the value of a vertically integrated rack-scale product.
AI Snips
Chapters
Transcript
Episode notes
Toshiba Drive Firmware Caused 2.7 Second IO Stalls
- Bryan Cantrill recounts Samsung buying Joyent to avoid huge public cloud bills and the pain of deploying hardware at hyperscale.
- At Samsung scale Dell substituted Toshiba drives causing 2.7s read stalls, revealing lack of integrated system control.
Component Substitutions Break Operability At Scale
- Hardware, firmware, and component substitutions beneath the OS create unresolvable operational pain when you don't control the full system.
- Third-party servers often substitute parts (e.g., different drive vendors) that break at scale and frustrate operators.
Broken BMC Sensor Led To Unnecessary 100W Fan Drain
- Bryan tells a customer story where a broken BMC temperature sensor made the controller index fan speed on CPU inrush current.
- Fans unnecessarily ran high for workloads that spiked current without temperature rise, wasting ~100W per server.