On Tue, Apr 03, 2001 at 06:40:09PM -0400, Russell Cattelan wrote:
> We haven't been able to duplicate this panic locally yet.
> Could you let us know what you are using for test machines
> client and server, number of processors in each, kernel version,
> linux distro version, speed of network between the machines.
> Finally what are you doing to "stress" nfs?
The server is a dual P-II 450MHz machine with an ASUS P2B-DS motherboard
and 256MB memory. It has a 3c905B ethernet card, leading to a Cisco
Cat 5000 switch. This machine is running RedHat Wolverine, with the
kernel upgraded to 2.4.3-XFS (i.e., stock 2.4.3 with the XFS patches).
It has an onboard AIC7890 SCSI controller.
It has 3 Seagate ST34502LW 4GB SCSI disks. One of these disks has been
formatted as XFS, and is exported via NFS.
The server has NFSv3 support enabled, and kernel NFSD.
The clients are 4 Linux boxes, each running RH6.2 or RH7.0. All have 100bT
ethernet hooked to the Cat 5000.
All of the clients mount the exported filesystem from the server. This
filesystem contains 1000 files of 3MB each. There is a "file list"
at the top level, which contains a list of all the files including the full
pathname.
Each of the clients runs a Perl script in a tight loop which does the
following:
- Grabs a random filename from this list
- stat()s and opens that file, and reads the entire contents
- close()s the file
One of the clients (a dual P-III 800MHz box) has a script that also
continuously writes files of size [4K .. 1MB]; it keeps at most 32 files
around, deleting old ones as it creates new ones.
This test setup looks weird, but it is designed to sort of simulate the
end use that this server is likely to get.
After about 18 hours or so of usage, the machine gets an oops. In this
time, the clients have each read about 38000 files each.
Ajay
|