Potentially show regressions when there are idle CPUs and a deep search isīeneficial. Potentially wins when domains are almost fully busy or overloaded -Īt that point searches are likely to fail but idle is not being agedĪs CPUs are active so search depth is too large and useless. Too high because a CPU is not reaching idle.ĭue to the nature of the patch, this is a regression magnet. The impact of this patch is related to workloads that have domains thatĪre fully busy or overloaded. To track this, the scan cost is subtracted from the estimated average

This means that we need toĬonsider the cost over all the wake-ups between consecutive idle periods. The overall goal is to not spend more time scanning for idle CPUs than New-idle balancing, which happens when we do go idle). Idle and age it (the existing value still makes sense for things like To fix this, introduce a separate average Is not at all idle, we'll happily continue using whatever value we did Time *for when we are idle*, that is temporally challenged. This is a partial forward-port of Peter Ziljstra's work first postedĪt Signed-off has been removed because it is modified but will be restoredĬurrently select_idle_cpu()'s proportional scheme uses the average idle ` (2 more replies) 0 siblings, 3 replies 19+ messages in threadĬc: Ingo Molnar, Juri Lelli, Vincent Guittot, Valentin Schneider, Sched/fair: Age the average idle time All of help / color / mirror / Atom feed * sched/fair: Age the average idle time 11:16 Mel Gorman

Vm horizon client latest version