The problem

How much useful work is lost when a scheduler interrupts a task sooner to make another task responsive?

What current research shows

EEVDF uses virtual deadlines and eligibility to decide which tasks should run and allows latency-sensitive tasks to request specific slices. The mechanism improves control over scheduling latency, but preemption and shorter execution intervals still carry context-switch and cache costs.

Where the evidence stops

There’s no universal best slice length. The right result depends on workload burst length, cache reuse, number of runnable tasks, and tail-latency requirements.

What Valen Systems is testing

Publish matched workload results showing throughput, p50/p95/p99 wakeup latency, context switches, cache behavior, and energy as the slice policy changes.

Sources

Linux scheduler: EEVDF