Chris Tyler píše v Po 17. 05. 2010 v 08:54 -0400:
On Mon, 2010-05-17 at 08:31 +0200, Dan Horák wrote:
> Chris Tyler píše v Po 17. 05. 2010 v 00:00 -0400:
> > On Sun, 2010-05-16 at 20:14 +0200, Till Maas wrote:
> > > On Mon, May 10, 2010 at 09:49:18PM -0400, Chris Tyler wrote:
> > >
> > > > The Fedora-ARM builds from around F8 though F12 were built on
> > > > koji farm in Europe. That was recently decommissioned, and
there's a new
> > > > koji farm being built here at Seneca by Paul Whalen. The plan is to
> > > > Dennis Gilmore's koji-shadow tool to follow updates on the main
> > > > so as updates are built for PCs they will also be built for ARM.
> > > > F13 should have consistent updates.
> > >
> > > Is there some public place, where this is documented? How many machines
> > > are used to build the build farm? Are these all Sheeva-style systems?
> > Lennert's farm had six Marvell dev boards with 2GB RAM if I understand
> > correctly. The Seneca farm will have 20 Sheeva-class builders once they
> > all arrive.
> I am a bit worrying whether the 512MB of RAM in the consumer products
> would be sufficient because I know that g++ can be very memory hungry
> when building Fedora packages. I experienced it on 64-bit platform
> (s390x), but 2GB of RAM (plus some swap) was too small there.
We ran some tests on some of the largest/slowest-to-build packages in
Fedora before the hardware was ordered, and 512M appeared to be
sufficient (and ARM machines with >512M are fairly rare, though this
will probably start changing).
My favorite is Scribus with it's very large functions/methods with many
variables inside, but the gcc developers fortunately made visible
progress to lower the memory consumption (at least in the 4.4 series).