ipid-debuggery (was Re: [e2e] ICMP & TCP segments with IP ID = 0?

[Francesco Potorti`]

Bob Braden:   
   I am confused by your claim.  Near the beginning of this thread,
   someone quoted RFC 791 as saying exactly that incrementing the IP ID is
   (always) required.  

I've been too concise.  Sorry about that.

Here is what rfc 791 says:
   | The identification field is used to distinguish the fragments of one
   | datagram from those of another.  

This suggests that the identification field is not useful for datagrams
where DF is set.

   |				     The originating protocol module of
   | an internet datagram sets the identification field to a value that
   | must be unique for that source-destination pair and protocol for the
   | time the datagram will be active in the internet system.  

Apparently we are talking about fragmentation.

Apart  from  that, I  think  that this  statement  has  not a  rigourous
meaning, because I think it is easy  to get into a situation where it is
impossible to obtain  a unique  value, so it should be interpreted  as a
guideline rather than a precription.

Also,  it could  be argued  that, in  order to  increase  probability of
uniqueness for  fragmentable packets, DF packets  should avoid consuming
sequence numbers  in systems where  a single counter  is used at  the IP
level.

Anyway,  I think  that increasing when possible  is a good thing.  Only,
systems that don't are not violating any standard, in my opinion.

Bob Braden:   
   And because it increases the system's robustness against failures caused
   by liberties (like turning off DF without resetting the ID)
   that boxes in the middle of the network might take.
   
Yes.

-- 
Francesco Potortì (researcher)         Voice: +39 050 315 3058 (op.2111)
Area della ricerca CNR - CNUCE         Fax:   +39 050 313 8091
via Alfieri 1, I-56010 Ghezzano, Pisa  Email: F.Potorti at cnuce.cnr.it
Web: http://fly.cnuce.cnr.it/          Key:   fly.cnuce.cnr.it/public.key