Re: Reducing number of i386 linux images
On Wed, Aug 22, 2007 at 08:20:22PM -0700, Steve Langasek wrote:
> On Wed, Aug 22, 2007 at 09:39:02AM -0400, Kyle McMartin wrote:
> > On Wed, Aug 22, 2007 at 09:18:44AM +0200, maximilian attems wrote:
> > > > I propose the following:
> > > > - Rename 686-bigmem to 686-pae. pae is more than support for much
> > > > memory, it includes things like NX.
> > > nack
> > > as already told on private channel to many pentium m out there are
> > > don't support pae
> > OTOH, there's little reason to ship a 486 *and* a 686 kernel, probably
> > easiest to just drop the 686 and rename 486 to generic or something.
> Is that based on benchmarks, or on an assessment that non-PAE 686 machines
> are uninteresting? I would think that if the performance benefit of 686
> over 486 is measurable at all, it's precisely the older systems that benefit
> most from having a separate kernel flavor, in terms of the hardware
> remaining useful.
I hope it will be based on benchmarks - I see no reason to drop any
flavours if there maybe a significant performance benefit for a
significant part of our userbase. I want as many people to run the
official kernel as possible and not to regress to the woody days
where any clueful admin built/ran their own.
In lieu of convincing benchmarks, the default answer should be to
*not* drop flavours. Without numbers, users are going to believe
(rightfully or not) that running Debian's kernel reduces the power
they get from their hardware. That factor is worth the slower build
times and extra disk space.
fwiw, a coworker is looking into the availability of a benchmark that
should be good for comparing pae/non-pae.
 whatever that means.