RE: avoiding tunnel fragmentation

Stefan Savage (savage@cs.washington.edu)
Thu, 29 Oct 1998 15:16:38 -0800

wont someone have to have the necessary state in order to reconstitute
the correct sequence number before handing it off to the destination
host?

Not necessarily. If the receiver doesn't send any data then there's
nothing for the sender to ACK so the ACK # stays the same. If there is
any two way data though you're correct that you'd need state. For that
reason I probably would prefer overloading checksum and urgent if this
needed to be done.

i very much hate to bring up the possibility, but it might be better
to explicitly label flows and recover necessary forwarding state based
on the label, rather than exploit a series of special cases to cram a
few bits here and there in the original packet

of course managing this state (distributed forwarding cache) is a
pretty severe pain pain in the ass

I'd argue that we should ignore the problem for now, but keep it in
mind. Doing it the "right" way (ie an MPLS-style forwarder) is
reasonable if this constitutes a problem that is of reserach interest to
us. Doing it the "wrong" way (hacking a few bits here and there) is
reasonable if we just need to eliminate fragmentation bias for our
measurements.