On Wed, 2009-02-04 at 16:01 +0100, Peter Robinson wrote:
>> >> So we should ignore OLPC that is deploying
100's of thousands all
>> >> running Fedora?
>> >
>> > Modest proposal: OLPC might benefit from running its own koji instance
>> > and effectively going the secondary arch route. -Os -march=geode, etc.
>> > Given how close it is to mainline x86 it's unlikely to have funky
>> > compilation failures, and it has to branch a non-trivial number of
>> > packages anyway.
>>
>> Actually the forked packages now are pretty minimal. I think there's
>> currently around a dozen, with the move the F11 that will be even
>> less. The major fork is the kernel but other than that most of the
>> forks are to slim down deps.
>
> Okay, ignore the bit about forked packages. How's the rest of the
> argument sound?
Sounds fine to me but I'm by no means an expert in compiler options
for different architectures :-) What sort of a win would we see
performance wise. Does it get done for all the packages or for just
things like kernel/glibc/openssl like it currently does for some of
the i386 packages?
Would be hard to know the performance win without trying it. But since
you're already bootstrapped, a mass rebuild would "only" be a few days
on a reasonably fast machine. It might be only a few percent of wall
clock performance, but depending how much RAM -Os saves, you might
thrash less, etc. (Roughly the same arguments apply for any global
compiler changes, for that matter.)
Doing this as a secondary arch would mean all packages would get
rebuilt.
I'm only kind of serious about this, since we really don't have
secondary arch support yet. But I think there's some merit in splitting
off sufficiently old (or subsetted) x86 from the real world.
Particularly if it means I can stop waiting for i586 kernels to build
just to get the i686 kernel I really want.
- ajax