A Day of Profiling on One Page
summary
Describes how Odnoklassniki profiles ~40,000 Java cloud services 24/7 using async-profiler wrapped with an eBPF layer and stored in an S3-compatible backend, and focuses specifically on the problem of processing and visualizing that data. Explains algorithmic techniques to compress gigabytes of profile data into a single HTML page (mentions LZ78) and engineering tricks in JavaScript to make browsers handle such large payloads efficiently.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 10:21Profiling data collection
- 22:23Profile data compression
- 33:02Unpacking and requirements
- 48:10Heatmap rendering
- 1:00:51Data structure optimization
- 1:10:31Summary
related talks
21mLeveraging Continuous Production Profiling for Providing Insights into...
01h 14mProfiling in Production
36mProfiling in the cloud-native era - Matthias Loibl
31mProfiling Formats and Why You Should Care
26mProfiling Mastery: Decode Your Data and Elevate Observability
46mProfiling your Java Application A Beginner’s Guide
31mProfiler: Profiling Code
45mJava Performance Analysis and Optimization