Autoscale — Pay When You Earn
summary
Designs architecture around unit economics and a single operational metric correlated with delayed revenue, then autoscale to that metric instead of guessing capacity. Recommends measuring revenue minus expense, shifting fixed costs to cloud using on‑demand, reserved, and spot instances (noting ~90% discounts), and adopting serverless (mentioning AWS free tiers: 1M Lambda requests and 400k GB‑s). Presents concrete tooling and numbers showing ~60% cost reduction and faster experimentation, and cautions about scale policies and cold starts.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:28Effective architecture
- 3:53Unit economics
- 7:10Ideal scaling rules
- 10:04Revenue and metrics
- 13:38Lean approach
- 15:22System design principles
- 18:37Single responsibility
- 22:11Hardware vs cloud
- 24:46Amazon instance types
- 27:43Spot instances
- 31:31CPU credits management
- 34:59Serverless advantages
- 37:12Serverless pitfalls
- 41:28Serverless risks
- 43:37Cloud recommendations
- 44:18Minimizing costs
related talks
31mAutoscaling Apps without Supersizing Costs
31mAutoscaling: the why, what, how & when to help you implement it
39mPrinciples of Automatic Application Scaling in AWS
26mScaling K8s Nodes Without Breaking the Bank or Your Sanity
18mResource allocation -- The usuals and beyond
35mBDMAAS: SMART AND BUSINESS-DRIVEN AUTOSCALING FOR LARGE-SCALE CLOUD BASED IT S...
42mHow We Saved Cloud Budget by Using Autoscaling
33mHow we do cloud cost efficiency at Notion