Messages in this thread | | | Subject | Re: [RFC PATCH 08/14] sched: normalize tg load contributions against runnable time | From | Peter Zijlstra <> | Date | Mon, 20 Feb 2012 17:10:19 +0100 |
| |
On Fri, 2012-02-17 at 04:32 -0800, Paul Turner wrote: > > Neither the comment nor the changelog mention this, it should, it should > > also mention why it doesn't matter (does it?). > > It doesn't and it should. Although I'll take the liberty shortening > it a little to something
For the in-code comment that's fine, it would be good for the changelog to have the entire story though.
> like "unfortunately we cannot compute > runnable_avg(tg) directly, however, XXX is a reasonable > approximation."
Yeah, not easily done indeed, you could compute a corrective term if you'd have something like the avg and variance of runnable over the various CPUs, but those too suck to have to compute.
| |