translating fragments

Robert L Ullmann <> Wed, 09 February 1994 01:44 UTC

Received: from by IETF.CNRI.Reston.VA.US id aa25310; 8 Feb 94 20:44 EST
Received: from CNRI.RESTON.VA.US by IETF.CNRI.Reston.VA.US id aa25306; 8 Feb 94 20:44 EST
Received: from by CNRI.Reston.VA.US id aa23726; 8 Feb 94 20:44 EST
Received: from by (5.65c/Spike-2.1) id AA00828; Tue, 8 Feb 1994 20:37:15 -0500
Received: by (5.65c/Spike-2.0) id AA00469; Tue, 8 Feb 1994 20:25:09 -0500
Precedence: bulk
Return-Path: <ariel>
Received: by (5.65c/Spike-2.0) id AA00457; Tue, 8 Feb 1994 20:25:07 -0500
Date: Tue, 8 Feb 1994 20:25:07 -0500
Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: Robert L Ullmann <>
Message-Id: <>
Subject: translating fragments


It is very good to get detailed comments; thank you.

I have a number of related comments, but carrying on an exchange
re multiple points is difficult, so I'll pick one to start with,
and bring the others back up as I do the close review of the doc with your


Translating fragments/segments without reassembling them would be
really nice; IPv7 suggested that at one point, butin the process of
doing trial implementation, I found a "little" problem: just how do
you figure out the offset in the resulting datagram?

Note that IPv4 and CLNP have this not-quite-stated assumption that the
recving host doing reassembly gets the "offset 0" fragment/segment
with a N-layer header that is THE SAME SIZE as the header was when
the datagram was fragmented by some router (or router*S*!) 

THis assumed axiom doesn't hold in the presence of different network
headers. I don't know how to solve this so as to allow translating
fragments; I'd love to have someone find something I've missed.

Note that it doesn't even hold for all-native-CATNIP, since the IPv4/CLNP
axiom doesn't apply; CATNIP forwarders can and will change the
header size when it pays to do so. 

The offset (IMHO) should have been defined as offset from the TPDU
start; an open issue is whether CATNIP-native should do this (I think it
should) but since IPv4/CLNP define it as NDPU-start relative, the 
translation problem still exists.

Ideas anyone?