High perf architecture at Stack Overflow
summary
Stack Overflow runs on nine web servers serving about 800 million pageviews per month with four beefy SQL machines, heavy Redis caching and a three-node ElasticSearch tag index. Performance work follows a tight build→measure→analyze loop using MiniProfiler and stored traces and frequent deployments (roughly three to four per day). Specific wins: a tag-query path drops from ~100ms to <10ms and a GPU plan targets <1ms, while compiler changes, pooled mutable strings and pragmatic inlined SQL cut allocations and GC pauses.
AI-written summary · about
chapters (17)
- 0:00Start
- 0:59What Is Architecture
- 1:57Utility
- 3:47What Is Performance
- 4:27The Scientific Method
- 5:28What Kind of Results Can You Get with this Methodology
- 8:55Servers
- 17:25Mini Profiler
- 18:35Sequel Database
- 23:05Documentation
- 27:41Garbage Collector
- 29:42Garbage Collection
- 33:38Resource Files
- 35:36Extending the Compiler
- 43:44Conclusion
- 45:28Hire the Right People
- 46:06New Silver Badge
