The SRE’s Crystal Ball: Predicting System Performance with...
summary
Uses Little's law and the Universal Scalability Law to move from firefighting to capacity predictions, with concrete numbers: 180 rps, 50 ms service time, 4 pods -> utilization 2.25 so 8 pods still yield 1.125 and 10 pods hit 0.9. Demonstrates a single-file Rust service (work/contention/coherence endpoints), load tool "work", and an R package fit that yields alpha=0.097 (contention) and gamma≈15, then extrapolates hardware scaling to cores (warning: noisy data and >2x extrapolations are unreliable).
AI-written summary · about
chapters (11)
- 0:00SRE performance prediction
- 1:43Queuing theory for SREs
- 5:20From firefighting to capacity planning
- 6:24Queuing theory basics
- 11:16Little’s Law in real scenarios
- 16:38Universal Scalability Law
- 22:19Live USL demo
- 31:40USL for bottlenecks
- 33:45USL limitations
- 34:27Key takeaways
- 37:17Q&A
related talks
43mThe Math of Scalability
01h 01mLittle things and big things: lessons for building large systems.
55mScaling out operational research with Rust
43mScalability Intro – Shyam Jeedigunta, Google & Bob Wise, AWS (Any Skill Level)
47mMonolith to Microserv. at 20M+ requests per second: Latency and Scale Challenges
34mDistributed Systems - What Can Go Wrong and How to Get It Right
46mPerfect Scalability
20mBetter Resilience Through Self Tuning