Re: Cache queue_congestion_on/off_threshold

[Jens Axboe <axboe@suse.de>]

On Thu, May 06 2004, Andrew Morton wrote:
> "Chen, Kenneth W" <kenneth.w.chen@intel.com> wrote:
> >
> > >>>> Jens Axboe wrote on Wed, May 05, 2004 11:43 PM
> > > On Wed, May 05 2004, Andrew Morton wrote:
> > > > Jens Axboe <axboe@suse.de> wrote:
> > > > >
> > > > > Do you have any numbers at all for this? I'd say these calculations are
> > > > >  severly into the noise area when submitting io.
> > > >
> > > > The difference will not be measurable, but I think the patch makes sense
> > > > regardless of what the numbers say.
> > >
> > > Humm dunno, I'd rather save the sizeof(int) * 2.
> > 
> > Strictly speaking from memory consumption point of view, it probably comes
> > for free since sizeof(struct request_queue) currently is 456 bytes on x86
> > and 816 on 64bit arch.  The structure is being rounded to 512 or 1024 with
> > kmalloc.
> 
> That's a good argument for creating a standalone slab cache for request
> queue structures ;)

Precisely, it's definitely not a good argument to keep loading it. I'll
do that.

-- 
Jens Axboe