CONCEPT Cited by 1 source
Serverless billing definition¶
Definition¶
Simon Willison's crystallization of "serverless":
"Serverless means per-request billing is my new favorite definition of that term. I'd previously been using 'scale to zero' but that always requires extra explanation."
(Source: sources/2022-07-11-highscalability-stuff-the-internet-says-on-scalability-for-july-11th-2022.)
The load-bearing consequence: the thing you actually pay for is not having to pay for idle. A 24/7-running cluster that happens to expose an event-triggered API surface is not serverless in any cost-meaningful sense. "Per-request billing" subsumes "scale to zero" (if you're not getting requests, you pay zero) and is easier to reason about at billing-audit time.
Operational consequence: Aurora Serverless v2 debate¶
Jeremy Daly argued Serverless Aurora v2 is not really serverless precisely because it can't auto-pause — idle instances still cost money, breaking the "pay for what you use" promise. The missing Data API and inability to scale compute to zero were named as the two fundamental misses. (Source: Serverless Aurora v2 still isn't serverless.)
Counterexample: Brian LeRoux on Lambda costs¶
- AWS Lambda economics: "$1 for 6M invocations. The whole point is stateless on-demand to achieve 100% utilization."
- "Provisioned server capacity is not 100% utilization. We over provisioned them to allow for spikes."
- "Comparisons where the Lambda is being invoked 24/7 is totally disingenuous. A cluster of servers running 24/7 mostly idle is not the same thing. Statelessness is a feature. It enables 100% utilization."
Pattern¶
Per-request billing + stateless-on-demand = utilization is priced at the invocation level, not the provisioning level. Traditional servers are over-provisioned for peak; serverless bills the actual peak-time work. For sub-1000-RPS workloads, this typically means 90%+ cost reduction versus an always-on equivalent (cf. thdxr's $4,000/mo EKS → $70/mo Lambda migration).