Performance in the Wild
summary
Relevance optimizes a Clojure JSON web service by building a stochastic latency model, running controlled load tests and profiling, then iterating small changes. Results move from 75 rps → 1500 rps (about 20× throughput) with 99% of requests under 10 ms at 1500 rps. Tooling is JMeter + YourKit for load and profiling. A surprise: syslog logging = 80% of request time, fixed with a thin C wrapper + Java bridge invoked from Clojure. Nightly automated tests catch config regressions and long‑run failures.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:05Craig Andera
- 2:00Performance Related
- 3:10Model Based Optimization
- 4:02The Problem
- 5:55Models
- 9:35Model Generation
- 15:25Benchmarking
- 19:50Tools
- 21:15Analysis
- 23:45Recommendations
- 24:52Optimize
- 25:43Indirection
- 26:17Start Early
- 27:20Relevance
- 27:55Where we started
- 30:30Average latency
- 32:39The Goal Line
- 37:24Be deliberate
