Scheduling, visually
Every scheduler in computing — CPU, cluster, disk, network, language runtime — is making the same four trade-offs. Fairness against throughput. Latency against utilisation. Predictability against adaptivity. Simplicity against optimality. Once you can see the trade-off, you can predict the failure mode.
- Linux replaced its CPU scheduler and nobody noticed
Why the kernel threw out sixteen years of CFS, and what the new scheduler lets you ask for that the old one did not.