Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!rutgers!elbereth.rutgers.edu!hardees.rutgers.edu!patterso
From: patterso@hardees.rutgers.edu (Ross Patterson)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: more IP (was Re: Re-fragmenting IP Datagrams)
Message-ID: <Mar.13.10.10.41.1989.4122@hardees.rutgers.edu>
Date: 13 Mar 89 15:10:43 GMT
References: <4014@ingr.com> <4537@jpl-devvax.JPL.NASA.GOV>
Distribution: usa
Organization: Rutgers Univ., CCIS
Lines: 14

>Anyway, one concept they have come up with is to 'trick' IP into thinking
>that a bunch of datagrams are actually datagram fragments (by forcing the
>appropriate fields in the IP header).  They expect that IP on the receiving
>end will simply recognize the things as fragments, reconstruct, and hand
>the fully formed datagram off to the application.

I suppose you've already considered the fact that one of these 
"super-datagrams" might have several sets of TCP or UDP headers in them, due
to the conglomeration of "fragments" that the receiving IP will perform? I 
expect that will cause most TCP/IP implementations to become deathly ill. 

Ross Patterson
Rutgers University
Center for Computer and Information Services