I received a patch from Sistina (thanks!) which does indeed fix the
oops occuring on the server. The patch apparently cleans up concurrent
usage of a _private buffer_head data field by ext3 and lvm. It does
*not*, however, alleviate the iozone error - that still occurs with
different semantics: iozone now exits cleanly after the i/o error is
displayed, NFS stays usable.
Trond: Is there anything I can do to get error messages/traces out of
the nfs layer? Or do analysis of the server (as it's fully usable
during/after the iozone run). Syslog (in full *.*) shows
nothing. Kudos for the patch btw - w/out I can't get write performance
better than 1.5MBps, with the patch it's an even 7.something MBps.
The iozone error reported below is 100% reproducible between those two
boxen.
Since the original report I've moved all mentioned machines to 2.4.18
(final, not -rc4) including Sistina's cleanup as mentioned above.
On a half-related note: I'm not subscribed to the nfs list and can't
subscribe to it without bending over backwards (the mail server
doesn't accept the confirmation mail targeted at it from an "dial-up"
(rather: dynamic-ip) account). No, the obvious simple solution is not
really viable.
Joachim Breuer <jmbreuer@gmx.net> writes:
> [1.] One line summary of the problem:
>
> During performance test (using iozone) against nfs server with
> kernel 2.4.18-rc4 + linux-2.4.18-NFS_ALL.dif (provided against
> 2.4.18-rc2, went in cleanly) from nfs.sourceforge.net + Sistina's
> LVM2 beta1.1, Oops occurs.
>
> [2.] Full description of the problem/report:
>
> Client and Server running same aforementioned kernel, NFS parameters
> set to rsize==wsize==4096; udp. Exported FS is an ext3 on a LV on
> two PVs (linear/segmented, not striped), one is a complete SCSI disk
> (sda), the other is a partition of an IDE disk (hda5). "./iozone -ac
> -R -n 256m" showed results for write/rewrite/read/reread; showed
> "Error writing block at 219414528", "write: Input/output error",
> "iozone: interrupted" before random read result. Result "never"
> comes up (not in 15+ minutes), iozone appears hung (Ctrl-C won't
> break it, kill -9 does).
>
> Checking at the server shows oops detailed below, and high load
> (around 8, stays up even after iozone was killed). No processes on
> server show responsible for the load average in "top".
>
> Same iozone against other NFS server works (more than one iteration,
> this NFS server will be called "reference server" from here on). The
> reference server is different hardware (see next paragraph and
> [7.7]), esp. different ethernet controller. Reference server does
> *NOT* employ LVM; but *does* also use ext3.
>
> All boxen X86, server is a Celeron 900 (PIII-Core-based); client is
> a Pentium III 850; running identical kernels. Reference server is a
> dual Pentium Pro 200 SMP, using the same kernel modulo SMP turned
> on. More info on hardware available on request.
[Rest of original problem report, including setup/machine description,
in <m3zo1q8kr2.fsf@venus.fo.et.local> posted on Sun, 03 Mar 2002 03:09:53 +0100]
So long,
Joe
-- "I use emacs, which might be thought of as a thermonuclear word processor." -- Neal Stephenson, "In the beginning... was the command line" - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Mar 07 2002 - 21:00:47 EST