Name lookup timeouts, redux -- IP defrag?

[dsb@goldfinch.cs.duke.edu (Scott Bigham)]

As you may recall, I recently reported problems with name lookup
(gethostname(), nslookup, etc.)  After some experimentation, I've
discovered that the critical variable is the size of the nameserver
reply packet -- in particular, its size relative to the MTU of the SLIP
connection.  For instance, the reply packet from the campus nameserver
for www.cygnus.com is 286 bytes long.  With an MTU of 256, this lookup
consistently fails; with an MTU of 350, it consistently succeeds, but
other lookups with longer reply packets consistenly fail.

This suggests a bug in IP defragmentation; the nameserver reply packet
appears to be getting dropped if and only if it gets fragmented.  Has
anyone else observed a similar problem?  Is anyone else successfully
running a SLIP connection with an MTU in the neighborhood of 256?

						-sbigham