[Samba] cifs problems with multiple linux boxes hitting W2K shares, cifs-1.20c-2.4, linux 2.4.27 kernel

[Joe]

A pair of clients hitting the server can't coexist.

We have a cluster of PowerEdge webservers running smbfs mounts.
I switched one box to cifs, 5 shares pointing at a W2K,
Monday 11 Oct. The box ran fine all week.
I ran load tests at 50 to 100 times the expected load
on several machines. I did not try those tests in parallel.
It's a moderate use production system.

This Monday I switched another box in the cluster to cifs.
Immediately both boxes began spewing errors like:
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: No response buffer
Oct 18 11:35:42 dlib2 last message repeated 2 times
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: Error -104 sending data on socket to server.
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: Error -32 sending data on socket to server.
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: Error 0xffffffe0 or on cifs_get_inode_info in lookup
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: No response buffer
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: No response buffer
Oct 18 11:35:42 dlib2 kernel:  CIFS VFS: Send error in Close = -11

Stats also shows: 2156 session 1688 share reconnects

One box finally gave up. I could not get apache to let go of the cifs
mounts. When I did a shutdown the box hung and I had to do a power cycle.

I switched the second box back to smbfs and now the first box is again
happily running cfis.

Any hints above how I can get both boxes to talk to the W2K server
at the same time? The W2K event logs didn't pick up any errors.

Thanks, Joe Edwards
UW Libraries