> I thought TCP was designed to guarantee error-free delivery of a
> packet.  Shouldn't TCP cause the packet to be resent if the IP
> checksum is wrong?

There's really no way to *guarantee* correct delivery of a message,
although in_cksum makes a really good try of it.  It all comes down to
probabilities.  In_cksum does a pretty good job of making sure that
one or two one-bit changes in a packet will be detected, but if you're
getting many bits changes in a packet from time to time, there's a
decent chance that a mutated packet will slip through occasionally.

A sixteen bit checksum isn't big enough to make this *really*
unlikely.  A 32-bit checksum is quite a lot better.  Running MD5 over
the message would probably be better still, but who has time?  I
notice failures about once every two weeks.  I'd like it if sup could
catch those.

			       _MelloN_
--
Ted Lemon							 mellon@vix.com
+1 415 567 7999