A Practical Look at Performance Theory
summary
Uses Little's Law and the Universal Scalability Law to turn queues into concrete predictions: a single-threaded web server shows a hockey‑stick response curve and crosses an SLA near 5500 rps (4,500 is safe, 5,000 already risky). Describes concrete mitigations—code optimization, batching to cut variance, controlled delay and last‑in‑first‑out serving used at Facebook—and warns that closed‑system load generators can understate production latency because synchronized clients hide open‑system behavior.
AI-written summary · about
chapters (13)
- 0:00Performance modeling intro
- 1:32Standard performance approaches
- 3:41Single server modeling
- 7:00Queuing delay vs response time
- 12:55Improving response time
- 17:12Proactive response time
- 19:20Closed vs open queuing
- 23:22Response time in closed systems
- 28:14Load simulation limits
- 30:03Scaling a cluster
- 33:22Universal scalability law
- 36:16Modeling with empirical methods
- 38:02Closing & Q&A
related talks
30mQueueing Theory, In Practice: Performance Modelling in Cloud-Native Territory [I]
06mPractical performance theory
46mQueueing Theory in Practice: Performance Modeling for the Working Engineer
47mPractical Performance Analysis
45mThe Universal Laws of Performance
45mUnderstanding Microservice Performance
30mThe Queue Continuum: Applied Queuing Theory
42mStop Rate Limiting! Capacity Management Done Right