Re: stochastic fair queueing in the elevator [Re: [BENCHMARK] 2.4.20-ck3 / aa / rmap with contest]

[Jens Axboe]

On Tue, Feb 11 2003, Jason Lunz wrote:
> [e-mail] said:
> >> By all means, do the same thing with disk i/o. It's been a smashing
> >> success with packet queueing.
> > 
> > Well, that's the point.
> 
> Yes, what you've done with cbq is great. What I was referring to,
> though, is the user configurability of network frame queueing. It's
> possible to do really complex things for very specialized needs, yet
> also easy to put in a simple tweak if there's just one type of traffic
> you need to prioritize.  It'd be nice to have that kind of
> configurability for unusual i/o loads, and the arbitrary queue stacking
> is a whole different beast than having a couple of tunables to tweak.

That is indeed the goal. We'll see how much is doable within the 2.6
time frame, though.

-- 
Jens Axboe