Dave-
[warning long on data, short on answers]
i've got a similar setup on a uniprocessor linux box serving to 3 IRIX
6.5.12m clients. i have not seen that behavior, but the nfs server is
hardly loaded. i did use all of trond's nfs patches from
http://www.fys.uio.no/~trondmy/src/2.4.6/linux-2.4.6-NFS_ALL.dif
a simple test seems to work okay:
irix-nfs-client$ dd if=/dev/zero of=irixnfstest bs=1048576 count=500
xfs-nfs-server$ uname -a
Linux xfs-nfs-server 2.4.7-pre3-xfs #7 Thu Jul 5 20:38:47 EDT 2001 i686
unknown
xfs-nfs-server$ uptime
11:18am up 35 days, 14:27, 1 user, load average: 0.00, 0.00, 0.01
now for the interesting part. i DO see similar behavior with a stock
2.4.3 kernel exporting only ext2 filesystems. i have seen both the
"burstiness" and the NFS server hiccups. this nfs server is heavily
loaded, and serves a software raid 0. there are 96 linux 2.4.x nfs
clients attached to this nfs server. this server does not have trond's
patches applied.
ext2-nfs-server$ uname -a
Linux ext2-nfs-server 2.4.3 #3 SMP Thu Apr 5 18:49:58 EDT 2001 i686
unknown
ext2-nfs-server$ uptime
11:26am up 8 days, 18:07, 6 users, load average: 4.31, 4.47, 4.47
i was thinking it had more to do with me eepro100 drivers screwing up
than anything. because i would get "eth1: card reports no resources."
messages preceeding those nfs errors. i never really worried about it
because its a production machine so i couldnt do any real testing on it,
furthermore it seems to heal itself after a little while.
also, i am currently testing an smp dell poweredge 2500 box with a large
dell hw raid volume, so if that matches your hardware i may be able to
see if i can reproduce your problem?
hope this helps or at least doesn't hurt. ;)
rob.
On 10 Aug 2001 10:47:21 -0400, Dave Sill wrote:
> I'm running 2.4.7-pre3-xfs checked out from CVS on July 11 on a
> 2-processor Dell server. I'm serving a 200 GB hardware RAID XFS
> filesystem to a couple of Origin 200 servers. (I had to apply a small
> patch to the kernel to work around an IRIX NFS bug having to do with
> getting the current working directory.)
>
> When I'm writing to the file system via NFS, performance is very
> bursty--it'll copy 200-300 MB pretty quickly, then stop for 2-3
> minutes--during which any attempt to access the RAID, even locally,
> hangs.
>
> The IRIX systems also frequently report that the Linux NFS server
> isn't responding. For example:
>
> Aug 10 08:41:15 6A:daacl unix: NFS server daacsnfs not responding still trying
> Aug 10 08:41:39 6A:daacl unix: NFS server daacsnfs ok
> Aug 10 08:42:13 6A:daacl unix: NFS server daacsnfs not responding still trying
> Aug 10 08:42:18 6A:daacl unix: NFS server daacsnfs ok
> Aug 10 08:42:24 6A:daacl unix: NFS server daacsnfs not responding still trying
> Aug 10 08:42:24 6A:daacl unix: NFS server daacsnfs ok
> Aug 10 08:45:11 6A:daacl unix: NFS server daacsnfs not responding still trying
> Aug 10 08:45:15 6A:daacl unix: NFS server daacsnfs ok
>
> (daacsnfs is the Linux box, daacl is an IRIX box)
>
> Any ideas about what's wrong?
>
> The Linux server also has a 200 GB e2fs filesystem that doesn't
> exhibit the same behavior.
>
> -Dave
|