re: some further musings

[sytek![e-mail] (Keith Mc Cloghrie)]

Art,
 
> If I understand your "report-N-fragmentations", then I am concerned about
> the state that must be kept.  If the state is only in the end system, then
> state must be kept for each IP source.  If state is kept in the router,
> then state must be kept for each IP source/destination pair.
> Also, would the report be generated for N successive fragmentations
> (probably not), N fragmentations within M packets, N framentations
> in time T, or ???

Yes, this requires that state be kept, but it requires no more
than 1063 does.  As for doing this in routers, I had not thought of
that before reading Jeff's suggestion for sending ICMP messages back 
from the last-hop router.  I agree that keeping this much state in 
routers is bad, and to be avoided.  

It was indeed for "N successive fragmentations".  I suppose the fact 
that you say "probably not" means that I explained it badly.  The 
intent was to send back one ICMP message for EACH of the next N 
fragmentations.  Given the assumption that any fragmentation 
is harmful, what I was looking to do was to trigger the ICMP 
message IMMEDIATELY on the very next receipt of a fragmented 
datagram.  My reason for specifying N (rather than 1) was so 
that even though some of the ICMP messages get lost on the return 
trip, there's a high probability that one of them gets through.  
On further thought, I guess it's still not guaranteed that at least 
one of them gets through, for which case I'm not sure how to recover.  
I'll have to think more about that.

Keith.