[e2e] Question on "identification" field of IP header

[Ramesh Shankar]

If the "Don't fragment bit" is set in the IP header, what purpose does 
the "identification" field serve? Why can't I simply put 0 for this 
field in such a case? I remember coming across some e-mail chain in one 
of the mailing lists (TCP-IMPL, e2e, TSVWG) about this issue and the 
interaction with NAT. But I am not sure what came out of that discussion.

Also, some of the "OS fingerprinting" techniques in the networking world 
apparently use the IP ID field to identify various operating systems. 
Apparently some operating systems (or protocol stacks, if you prefer) 
don't increment this by 1. Why would anyone want to do this? Wouldn't 
this cause possibles issues wrt to wrap around in the possible case of 
fragmentation? One could always argue that with high speed Ethernet, 
that anyway would be a possibility, but I would like to understand the 
reason behind such a case.

Thanks for your time,

S.R.