infrastructure
Scale to Zero Reaches the Physical Server
A tinyFaaS scheduler powers single-board-computer workers off between bursts, cutting mean draw while moving delay into the tail.
Summary
A tinyFaaS scheduler powers single-board-computer workers off between bursts, cutting mean draw while moving delay into the tail.
The platform treats node power state as part of serverless worker selection. On a physical cluster, the authors report 39.6% lower mean draw for a bursty trace and 46.7% lower draw for a sparse one, without a detectable median-latency change on the bursty case. Invocations arriving when every worker is off can wait about a minute for boot.
Why it matters
A tinyFaaS scheduler powers single-board-computer workers off between bursts, cutting mean draw while moving delay into the tail.
Limits and context
No additional limitation was separately recorded.
Key claims
A tinyFaaS scheduler powers single-board-computer workers off between bursts, cutting mean draw while moving delay into the tail.
Evidence: source-2026-11-11-021
Sources
- arXiv:2611.14026v1arXiv / Cornell University · primary research
Corrections
No corrections have been recorded for this story.