Re: [patch] "HT scheduler", sched-2.5.63-B3

[Linus Torvalds]

On Thu, 6 Mar 2003, Martin J. Bligh wrote:
> 
> It would be nice if we had a "batch-job-able" simulation of this situation,
> to do more accurate testing with ... can anyone think of an easy way to
> do such a thing? "waggle a few windows around on X and see if it feels 
> better to you or not" is kind of hard to do accurate testing with.
> Of course, the simulation has to be accurate too, or it gets stupid ;-)

It should be possible to use the same approach that the other latency
testers use (ie contest&co), by just generating some specific background
load (say, different mixtures of X clients and plain compute-bound things 
like kernel compiles).

Ie measure directly exactly what we're interested in: the latency of some
general X request, under different loads.

		Linus