A small scheduler patch improves desktop responsiveness
Summary
LWN puts into context a much-discussed Linux patch that treats interactive loads more fairly against large build jobs through automatically formed task groups. Perceived speed depends more on latency spikes than on maximum total throughput. Small kernel changes can have big effects when they target a central resource decision.
Ideas
- The scheduler can group processes by session instead of weighting every single task equally.
- CPU-intensive builds therefore crowd out interactive programs less.
- A small patch changes behaviour because it organises existing scheduler mechanisms differently.
Recommendations
- Measure responsiveness and total throughput separately under the same mixed load.
- Check scheduler changes against server, desktop and real-time profiles.
References
Links to the original source and the Web Archive open in a new tab.