The problem
What happens when a task becomes performance-critical before its utilization history catches up?
What current research shows
Linux documents PELT as an exponentially weighted history, with recent activity contributing strongly but not exclusively. The utilization-clamp documentation describes user-space hints as a way to overcome ramp-up delay for tasks whose deadline arrives before the signal reflects their new demand.
Where the evidence stops
A faster response can spend more energy or boost work that turns out not to matter. A hint is a request within system constraints, not a guarantee.
What Valen Systems is testing
Measure wake-to-completion latency and energy when demand changes abruptly, comparing historical signals with explicit workload declarations.