On Wed, Aug 13, 1997 at 09:52:18PM -0500, Allan 'Norm' Crain wrote:
> tcpdump output shows this (cesium is the ELC, obrien is a 386dx/40 that is
> acting as an NFS fileserver and boot server for Cesium)
> 
> 21:48:41.790000 cesium.idea-inc.com.542f6af > obrien.idea-inc.com.nfs: 1472 write [|nfs] (frag 1815:1480@0+)
> 21:48:41.800001 cesium.idea-inc.com > obrien.idea-inc.com: (frag 1815:1480@1480+)
> 21:48:41.800003 cesium.idea-inc.com > obrien.idea-inc.com: (frag 1815:1480@2960+)
> 21:48:41.800005 cesium.idea-inc.com > obrien.idea-inc.com: (frag 1815:1480@4440+)
> 21:48:41.800007 cesium.idea-inc.com > obrien.idea-inc.com: (frag 1815:1480@5920+)
> 21:48:41.800009 cesium.idea-inc.com > obrien.idea-inc.com: (frag 1815:936@7400)
> 21:48:46.200000 obrien.idea-inc.com > cesium.idea-inc.com: icmp: ip reassembly time exceeded [tos 0xc0]

Does other intensive file I/O have the same effect? Could you perhaps ktrace
the pwd_mkdb and see how much it is trying to write? Also, it looks like
you're using TCP for the NFS mount, does it work with UDP?

- Frank