[EMAIL PROTECTED] wrote on Thu, 25 May 2006 15:39 -0500:
> > Do you think it would be better to instead run with the
> > --enable-nptl-workaround option instead?
>
> Could you try that out? I am not sure why we had those options.. Perhaps,
> someone else who remembers can comment on this? (It is perhaps an x86
> option in any case!)
> It is worth a shot in the meantime...
> Sorry I forgot, what exactly was the problem that you are facing? Is it a
> performance problem or a functionality issue?
All it does is install the script pvfs2-server-stub as pvfs2-server,
with the real executable installed as pvfs2-server.bin.
The script does basically:
export LD_ASSUME_KERNEL=2.2.5
pvfs2-server.bin
to try to disable NPTL. I don't know if it would have any effect.
Lee: you could just put that "export" line in your startup script
before it starts pvfs2-srever and try again to see if it magically
fixes things.
We also have libc-2.3.2.so on our piv cluster, and are using RHAS3,
though the 32-bit x86 version, probably about the same vintage as
Lee's. I have had no end of problems with this machine and NPTL and
AIO in particular. Finally we ended up disabling NPTL by removing
/lib/tls to make some applications happier. Doing so also forced a
rebuild of some libraries that assume NPTL, including db4, so I have
a custom install of that for pvfs2. I also have to
--disable-aio-threaded-callbacks else AIO threads spin at 100% CPU;
it works with threaded callbacks, just with lots of CPU load.
I'm not about to advise Lee to go to these lengths, though. But
maybe getting rid of that buggy NPTL via the LD_ASSUME_KERNEL will
help.
-- Pete
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
