From Continuous Profiling to Continuous Optimization
summary
Production-ready open-source profiler called gProfiler (open source) continuously samples kernel and user stacks with low overhead and builds flame graphs so teams find code hotpaths and quantify CPU waste. One internal case replaces regex with a finite state machine and cuts a function from 40% → 9% regex hotpath dropping cluster CPU from 50% to 25% and cluster size from 10 to 4. A Django customer caches DB calls and shrinks 18.5% → 3.3% DB queries. Complementary product is an agent-based OS+runtime optimizer that claims typical cost reductions around 30–40% with no code changes on Linux.
AI-written summary · about
chapters (16)
- 0:00Introduction and speaker intro
- 1:03Why optimize applications
- 2:28Traditional optimization methods
- 3:11What is continuous profiling
- 5:17How continuous profiling works
- 6:44Benefits of profiling with flame graphs
- 8:28Introducing GProfiler tool
- 10:14Use case: Regex optimization
- 13:04Use case: Database query caching
- 15:11Introduction to continuous optimization
- 16:34Granulate’s optimization approach
- 17:16Example: Netflix resource allocation
- 19:23Granulate’s technology overview
- 21:27Connecting profiling to optimization
- 22:53Key takeaways and next steps
- 23:34Q&A and closing remarks
related talks
26mOptimize Performance with Continuous Production Profiling
06mOptimizing Performance - Continuous Production Profiling
01h 13mContinuous Profiling: The Missing Piece in Your Observability Puzzle
38mA Profile of Performance Profiling With pprof
01h 01mDebugging and Profiling .NET Core Apps on Linux
01h 32mBlazing Performance with Flame Graphs
48mProfiling is for Dev, Monitoring is for Ops – no more!
46mContinuous Profiling - the 4th Pillar of Observability