NFS client hangs copying data to file on server

[rettw]

I am seeing some strange behavior with NFS on gentoo-dev-sources 2.6.7-r12. I have a process that does a cat /dev/video0 > file.mpg where file.mpg resides on an NFS filesystem. Under kernel 2.4, this works fine and successive ls commands show the byte count going up each time. On 2.6.7, this process never works past 70 or so MB, it hangs at this point. I can control-c the cat and it starts again up to 70 or so MB. It seems to get better if i lower the bitrate that the file is written. Successive ls's show the file counting up. The server is running redhat-8.0 with a 2.4.21 kernel.

For a test I tried the same process to a different server running the same gentoo-dev-sources kernel and I see similiar behavior, except that the nfs client seems to batch the file writes. Ethereal confirms this - no traffic except for bursts of 20000+ packets every few minutes. Successive ls's only show the file growing at 30MB increments. The hang problem still exists though.

Using cp to move files back and forth works fine. Just catting exhibits the issue.

If i reboot the client back to 2.4.26 (gentoo-sources) everything seems to work fine. What could be going on here?

I am exporting the file systems with the sync option (async does the same thing) and I am using nfs-util-1.0.6 on both gentoo boxes. I also tried specifying rsize/wsize=8192 to no avail in mount command. I am using UDP not TCP and NFSv3 support is present in all kernels.

Thanks in advance,

Rett