Hi Jaroslaw,
> >>On Wed, Jan 04, 2006 at 12:30:39AM +1000, Michael Mansour wrote:
> >>
> >>>>do_IRQ: stack overflow: 284
> >>>> [<c01078a2>] do_IRQ+0x44/0x130
>
> Yep, that looks very much as a "side effect" of 4K stacks.
>
> The next SL 4.X i686 kernel will have this option changed (it
> requires disabling netdump patches).
>
> I would suggest to use SL4 for x86_64 for your file servers
> (providing you have hardware which can support it): 4K stacks are
> issue only on 32bit architecture.
Yeah, I gathered that from all the comments made in various other lists when
searched through google.
> > there's also a "dmapi" rpm in there which I haven't installed, I'm not sure if
> > I need to or not?
>
> Unless you have applications using DMAPI (Data Management API): no.
> (AFAIK there are not many such apps available for now ..)
>
> > I also noticed that there's a Dag Wieers package there:
> >
> > xfsprogs-2.6.13-3.rf.i386.rpm
>
> As far as I can tell 2.6.13-3.el.rf is just rebuilt of 2.6.13-3.rf
> without changes (looking at spec file)
>
> > And the last thing I guess is, would it be feasible to try a CentOS 4
> > contributed XFS kernel package? would it even work on SL4?
>
> Yes it will work.
> BTW: I looked little bit at CentOS XFS kernels hoping that they will
> have a newer codebase of XFS there .. nope: it seems that they just
> do the same as we (plus enabling 4K stacks) .. unless that changed
> recently ..
To me, it's quite dissapointing that this problem even exists and from the
multitude of results you get out of a google search, it's been around for ages
and hasn't really been considered something to permanently fix. I remember
reading on one of the google results that one guy from Red Hat commented that
"not many people are affected by this so it's not worth worrying about".
I spoke to SGI today also and they just said "grab the latest kernel from
kernel.org, recompile it using RH's config and enabled an 8K stack, but
there's no guarantees you won't have issues".
Either way, I've thought long and hard about this and have relunctantly
decided to get rid of the XFS filesytems and migrate them to ext3. For me,
these are production clusters, and I may just be causing myself more issues in
the long run if I steer too far away from mainstream (RH supported) kernels.
I've been with RH distributions since RH 4.2 (gee, is that 12 years?), but to
be honest, after this issue I'm starting to think I may be better off for
future builds to be looking at SUSE. I took a look at SUSE 10 about a month
ago and it was so similar to RHEL4 that a migration from RH would seem to be
seemless. SUSE also seem to support things I'm after right out of the box,
including XFS and php5. RH seem to be lagging behind, and in this
internet/technology world, that's not good.
Michael.
> Jarek
>
> __
> -------------------------------------------------------
> _ Jaroslaw_Polok ___________________ CERN - IT/FIO/LA _
> _ http://home.cern.ch/~jpolok ___ tel_+41_22_767_1834 _
> _____________________________________ +41_78_792_0795 _
------- End of Original Message -------
|
