xfs
[Top] [All Lists]

Re: fc3 and stacks

To: linux-xfs@xxxxxxxxxxx
Subject: Re: fc3 and stacks
From: Robin Humble <rjh@xxxxxxxxxxxxxxxx>
Date: Mon, 14 Mar 2005 23:46:06 -0500
In-reply-to: <4235F824.6070303@sgi.com>
References: <20050310232036.GA19295@lemming.cita.utoronto.ca> <4234E903.8010309@thebarn.com> <4235D44F.1020902@tippett.com> <20050314190915.GB9784@lemming.cita.utoronto.ca> <4235F824.6070303@sgi.com>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.4.1i
On Mon, Mar 14, 2005 at 02:46:28PM -0600, Eric Sandeen wrote:
>be sure.  But I've always meant to try to put together some diabolical 
>"supported" config on RHEL4 and file a bug - say nfs over deeply nested 
>volume managers, or something like that.

good plan.

>Seems like with RHEL3, stock kernel.org kernels don't play nicely.  I 
>hope this isn't the case with RHEL4, but I have not tested it.

I haven't found any problems running AS4 with generic 2.6 kernel.org
kernels. Or generic 2.4 or 2.6 kernels with fc2 or fc3 either
(+/- using dev instead of udev with 2.4 kernels and fc3/as4).

I believe that the new thread model (NPTL) was the biggest problem, but
glibc now seems very good at working out which kernels have which
features and making allowances. Maybe early fc2 had problems?

>But rebuilding the stock RHEL4 kernel, with tweaks to turn xfs back on, 
>maybe update xfs, and re-enable 8k stacks should be possible, if not 

Here's a first attempt.
As long as you're prepared to lose the RedHat crashdump stuff (at least
some of the 1500 series patches assume the 4k stacks imposed by patch200)
it seems to work ok.

Attached is a patch to re-enable 8k stacks and turn on XFS in a RHEL
kernel. ie.
  wget 
http://ftp.redhat.com/pub/redhat/linux/updates/enterprise/4AS/en/os/SRPMS/kernel-2.6.9-5.0.3.EL.src.rpm
  rpm -ivh kernel-2.6.9-5.0.3.EL.src.rpm
  cd /usr/src/redhat
  patch -p1 < rhel_as4_turn_on_8kstacks_and_xfs.patch
  rpmbuild -ba --target=i686 SPECS/kernel-2.6.spec

which builds, installs, and runs for me. I can mount XFS partitions,
and recompiled the kernel as a stress test, but I haven't pushed it
past that stage... so YMMV, no guarantees, I've probably missed
something, it will likely cause a you to be crushed under an infinite
pile of jellybabies, etc. etc. :-)

As an added tweak, I like to ^C the rpm rebuild after it has finished
all the patching and has just started the compile, and then
'make menuconfig' and turn on (at least) cpu specific optimisations
(eg. Xeon) and turn off generic x86. That makes a measurable difference
for speed in communicating between threads - faster context switching
or spinlock code I guess.

cheers,
robin

Attachment: rhel_as4_turn_on_8kstacks_and_xfs.patch
Description: Text document

<Prev in Thread] Current Thread [Next in Thread>