Pushing Git repos over the speed limit.
summary
GerritForge and partners report concrete speedups: using a Gatling extension to compress one year of traffic into an hour, tuning bitmap parameters cuts bitmap creation from 2.5 hours to 2 seconds, ref-table in‑memory caching removes repeated FS encoding, and packfile/preserve strategies avoid stall handles. An autonomous agent named Franco applies AI-driven reorganizations in real time across Gerrit, GitLab and GitHub Enterprise, producing up to 100× warp acceleration and roughly 90% lower CPU cost in their tests.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:07Performance as core feature
- 7:42Ref table and bitmap reachability
- 13:02Git performance tuning
- 16:59Bitmap generation speed
- 18:04Ref table encoding efficiency
- 19:26Research findings
- 21:11AI-driven performance agent
- 22:14Performance results with Garrett
- 23:59GitLab performance improvements
- 25:48Formula 1-level scaling
- 26:53Deployment and future plans
related talks
34mSponsored Session: Sustaining Git Performance Under Heavy Workloads- 34mSustaining Git Performance Under Heavy Workloads
47mSo you think you know Git: Advanced Git Tips and Tricks
27mB How GitLab scaled Git access with Go and what we gained from it
38mHow We Scaled GitLab For a 30k-employee Company
29mScaling Across the Enterprise
47mSource Control & Remote Team Collaboration
36mLessons from building GitHub code search