Serverless Systems That Won't Burn a Hole in Your Pocket
summary
Shows concrete levers: measure duration × memory × invocations to get cost (example: 200 ms × 1M = $166 + $0.20 ≈ $186), and use tooling to find sweet spots. Use arm runtime saves ~20% when compatible, move heavy one‑time setup into init to collapse cold starts, and choose memory to trade latency for throughput (one prime‑compute example runs 8× faster at 1024 MB for nearly the same price). Run Lambda Power Tuning or Compute Optimizer, use provisioned concurrency ~60% cheaper when utilized, batch and filter events, and prefer Step Functions direct integration to cut idle time.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 4:20Cost matters in serverless
- 5:22Lambda billing explained
- 11:11Memory latency vs throughput
- 15:23Optimizing cold starts
- 22:31Optimal memory allocation
- 25:18Reducing idle time
- 29:09Offloading requests
- 31:15Avoiding unnecessary invocations
- 33:21Direct integrations
- 37:19Cost optimization takeaways
- 39:06Q&A
related talks
42mServerless Systems That Won't Burn a Hole in Your Pocket
58mServerless: Five Key Things you need to Know
54mServerless beyond Functions
15mAlternative Ways To Think About Serverless
47mMoney-saving tips for the frugal serverless developer
41mServerless systems that won't burn a hole in your pocket
44mServerless Architecture on Production: Good, Bad and Ugly Parts
01h 00mServerless - reality or BS - notes from the trenches