Re: Kernel support for specific cpus
On Mon, May 14, 2001 at 06:19:50PM -0400, Raul Miller wrote:
> So far, I've not been able to turn up any benchmarks on linux kernels
> compiled for different cpus. There's stuff on kernel.org showing
> that years ago someone had plans for documenting such benchmarks, but
> that's the best that I've turned up so far.
>
> That leaves me with only conjecture and superstition (mine is that
> these speed differences aren't measurable for normal applications --
> otherwise someone would have documented them). However, conjecture
> and superstition aren't a suitable basis for a technical committee
> decision.
>
> Does *anyone* have any relevant, solid data that they can point us at?
>
> If there's no hard data -- neither positive nor negative -- I think we
> should adopt a wait&see attitude, expecting that [any day now] such
> benchmark results will turn up, or that we'll soon have the resources
> to conduct such benchmarks ourselves.
well, i think we have evidence of negative impact (bloated number and
size of kernel-image and kernel-headers packages - with associated
bandwidth costs and an inevitable increase in the number of CD-ROMs
required for a release), but no evidence of any positive impact.
in the absence of any benchmarks or other evidence showing positive
effects which outweigh the negative impact, we should go back to the old
practice of only having enough kernel-image packages to do the job of
installing debian, rather than compiling having of kernel-image packages
for an arbitrary assortment of cpu-specific optimisations.
if any such evidence turns up in future, the issue should be re-examined
at that time.
craig
--
craig sanders <cas@taz.net.au>
GnuPG Key: 1024D/CD5626F0
Key fingerprint: 9674 7EE2 4AC6 F5EF 3C57 52C3 EC32 6810 CD56 26F0
Reply to: