Task Scheduling for Multicore Embedded Devices
summary
Implements distributed weighted round-robin (DWR) inside Linux and measures it on a quad-core Cortex-A9 device running Ubuntu 12.04 / kernel 3.1.41. Claim: DWR yields near-perfect per-core fairness, visibly equalizing CPU shares versus CFS. Benchmarks: 150s video playback and database (random small I/O) show higher throughput and lower completion time; SunSpider JavaScript is mixed (single-threaded bias). Known drawbacks: DWR ignores migration cost and cache-miss penalties, so fairness gains come at potential latency and cache overhead; DWR is an improvement but not a final solution.
AI-written summary · about
chapters (8)
- 0:00Multi-core processor trends
- 1:54Advanced core architectures
- 4:44Linux scheduling challenges
- 6:12History of Linux schedulers
- 11:33Multi-core load balancing
- 18:55Distributed WRR algorithm
- 29:47Experimental results
- 35:21Conclusion and future work
related talks
44mReal-Time Linux on Embedded Multicore Processors
46mOptimizing Scheduler for Linux Gaming
35mOpening the Box: Diagnosing Operating-System Task-Scheduler...
32mImprove CPU Utilization by Designing a New Schedule Algorithm
46mOS Scheduling with Nest: Keeping Tasks Close Together on Warm Cores
38mFixing Real-Time Scheduler Throttling in the Linux Kernel
52mEffectively Measure and Reduce Kernel Latencies for Real-time Constraints
36mJourney of improving RT performance on Single CPU equipped PREEMPT_RT Linux