The Importance of Continuous Regression for HW & SW Development
summary
Ampere builds a TARS continuous regression stack that runs bare-metal provisioning on 32-core 3GHz hardware with 512GB RAM using Ansible, fironix benchmarks and perf-backed analysis. It ingests results into a database, computes OS health scores, flags high coefficient-of-variance or bimodal regressions, and correlates step changes to kernel or package updates. Sampling 20 large projects (≈200M LOC, 1.2M commits by 31K contributors in 12 months) surfaces firmware install failures and kernel-driven performance gains.
AI-written summary · about
chapters (36)
- 0:00Introduction
- 0:17About Ampere
- 1:30About Continuous Regression
- 2:51Hardware vs Software
- 3:29Infinite Problem Space
- 3:55Large Open Source Projects
- 4:27Continuous Change
- 4:37Maintain Quality
- 5:07Automated Regression System
- 5:55Open Source Partnership
- 6:47Three Phases
- 7:55Scheduling
- 8:07Test Sets
- 8:31OS
- 9:44Dashboard
- 11:36Boot Script
- 11:52Package Update
- 12:15Ansible
- 12:46Reproducible
- 13:08Test Execution
- 14:46Test Extensions
- 15:51Unstructured Compute
- 16:07Database
- 19:05Bimodal Results
- 21:09Pattern Matching
- 21:39Statistics
- 23:35Perf Charts
- 25:03Packages Versions
- 26:05Open Source Partners
- 26:57Contact Information
- 27:48Iftars
- 31:12Perf Data
- 36:28Load Generation
- 37:44Rack Deployment
- 38:18Testing Load Generation
- 40:11Customers and Partners
related talks
23mStop Performance Regressions Before Release Valkey’s CI Benchmarking
24mContinuously Integrating the Upstream Linux Kernel on Hardware
01h 00mContinuous Regression Testing for Safer and Faster Refactoring in C++
37mAutomated Testing and Continuous Integration/Delivery: Lessons Learned
43mOptimizing Full-Stack Sustainability in a Real World Data...
16mServiceLab: Preventing Tiny Performance Regressions at Hyperscale through...
22mFear of the computer
47mUsing Benchmarks for Tracking Regressions and Improving UI performance