Twitter’s Quest for a Wholly Graal Runtime
summary
Twitter builds an OpenJDK 8 with JVMCI and Graal, swaps Graal in for C2 on real Finagle services, and measures concrete wins: 11% CPU savings and 2.7% fewer scavenge cycles on a dedicated-instance test of the tweet service. Build-time/bootstrapping costs are visible (~15–30s bootstrap hit), several Graal bugs (bytecode snippets, byte-swap, stack-replacement) get fixed, a few intrinsics (string compare) are ported, and Graal runs in production with plans to expand.
AI-written summary · about
chapters (20)
- 0:00Intro
- 0:22Goal
- 4:30Bugs
- 9:50Compact Compact Strings
- 11:59finagle
- 15:45requests per second
- 16:37parallelgc
- 18:04growl
- 18:41Metadata
- 19:34C2 vs C9
- 20:15Why X is Right Ratio
- 21:08How Much Does It Cost
- 24:02Can We Save Even More
- 26:18We Are Happy
- 26:48Whats Not Announced
- 29:18Give It A Try
- 30:37Inlining
- 34:18Enterprise Crawl
- 36:08Can Graal run 100 in Java
- 37:37Other languages
related talks
50mTwitter's quest for a wholly Graal runtime
53mTwitter's quest for a wholly Graal runtime
54mTwitter's quest for a wholly Graal runtime
41mGraal: How to use the new JVM JIT compiler in real life
56mPerformance tuning Twitter services with Graal and Machine Learning
40mScala ♥ Graal
05mGraal
54mDeveloper's Introduction to GraalVM