On Wed, 18 May 2005 at 1:43pm, Gregory Brauer wrote
> Joshua Baker-LePain wrote:
> > Do you have a test case that would show this up? I've been testing a
> > centos-4 based server with the RH-derived 2.6.9-based kernel tweaked to
> > disable 4K stacks and enable XFS and haven't run into any issues yet.
> > This includes running the parallel IOR benchmark from 10 clients (and
> > getting 200MiB/s throughput on reads).
> >
>
> We first saw the problem after 5 days in production, but since then
> we took the server out of production and used the script
> nfs_fsstress.sh located in this package:
>
> http://prdownloads.sourceforge.net/ltp/ltp-full-20050505.tgz?download
>
> We run the script on 5 client machines that are running RedHat 9
> with kernel-smp-2.4.20-20.9 and nfs-utils-1.0.1-3.9.1.legacy and
> are NFS mounting our 2.6 kernel server. The longest time to OOPS
My clients are all RHEL3. I'll give the nfs_fsstress scripts a shot.
> has been about 8 hours. We have not tried the parallel IOR
> benchmark. (Where can we get that?)
http://www.llnl.gov/asci/purple/benchmarks/limited/ior/
> You didn't mention if you are using md at all. We have a
> software RAID-0 of 4 x 3ware 8506-4 controllers running the
> latest 3ware driver from their site. The filesystem is XFS.
> The network driver is e1000 (two interfaces, not bonded). The
> system is a dual Xeon. We upped the number of NFS daemons
> from 8 to 64. The nfs_fsstress.sh client mounts the servers
> both UDP and TCP, and our in-production oops likely happened
> with a combination of both protocols in use simultaneously as
> well. We've seen the OOPS with both the default and with 32K
> read and write NFS block sizes. The machine was stable for
> over a year with RedHat 9 and 2.4.20.
The server I'll be testing is dual Xeon, 2GB RAM, 2 x 3ware 9500-12
controllers running the 9.1.5.2 firmware and using the stock centos-4
drivers. (I see a huge performance hit using the latest 3ware driver set,
9.2.) The 3wares are doing hardware RAID5, and I do software RAID0 across
'em. Networking is the same -- 2 e1000s, not bonded. I'm running 256 NFS
daemons, well, just because.
--
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
