lkml.org 
[lkml]   [2008]   [Aug]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [Bug #11342] Linux 2.6.27-rc3: kernel BUG at mm/vmalloc.c - bisected
On Tue, Aug 26, 2008 at 12:09:46PM -0700, Linus Torvalds wrote:

> If you want the default kernel to support 4k cores, we'll need to fix the
> stack usage. I don't think that is impossible, but IT IS NOT GOING TO
> HAPPEN for 2.6.27.
>
> And quite frankly, if some vendor like RedHat enables NR_CPUS=4096 by
> default, they are totally and utterly crazy.

heh. *picks through Fedora changelog*

* Thu Aug 14 2008 Dave Jones <davej@redhat.com>
- Bump max cpus supported on x86-64 to 4096. Just to see what happens.

I never did get to find out unfortunatly, because of the security fiasco
in Fedora infrastructure the last week or two.

> But some SGI-specific binary that is meant for SGI machines only, and has
> been extensively tested with the setup used on SGI machines is a different
> thing.

Every extra kernel image a distro vendor ends up shipping has an associated cost.

* build time: It currently takes about 2 hours for a set of Fedora RPMs.
For RHEL it'll be even worse due to the extra archs).
Killing off -smp specific builds was a big win for us in this regard.
Adding extra flavours is always painful.

* diskspace (distro kernels aren't small. With the associated debugging symbols,
they take up a shitload of disk space really fast).

* Having everyone running the same kernel makes it much easier to test/debug.
Our QA guys hate adding extra columns to their test matrix.

But yes, for this to be even remotely feasible, there has to be a negligable
performance cost associated with it, which right now, we clearly don't have.
Given that the number of people running 4096 CPU boxes even in a few years time
will still be tiny, punishing the common case is obviously absurd.

Dave

--
http://www.codemonkey.org.uk


\
 
 \ /
  Last update: 2008-08-26 21:41    [W:0.351 / U:0.892 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site