Meta-issues: On the deprecation of the fragmentation function

[Fernando Gont <[e-mail]>]

Folks,

I wanted to comment on some met-issues regarding the deprecation of the
IPv6 fragmentation function.

** On the motivation of deprecating the fragmentation function **

So far (and without having read Ron's recent I-D -- shame on me), it
looks like the main two reasons for deprecating the fragmentation
function are:

1) The inability of middle-boxes to parse past the first XXX bytes of a
packet

2) Unavailability of the connection-id (five-tuple) in the non-first
fragments.

Regarding "1)", I believe that deprecating fragmentation is not really
the right solution. If anything, one could require the entire header
chain to be within the first XXX bytes of a packet (as a former version
of draft-ietf-6man-oversized-header-chain did). Besides, if we're going
to deprecate the fragmentation function because of this, then we should
also deprecate all extension headers, because they might lead to the
same issue.

Regarding "2)", IPv4 doesn't have the connection-id in the non-first
fragments, either. So whatever middle-boxes are doing, they should/could
do it in the same way as they currently do for IPv4.


** On the impact on applications **

It has been stated that fragmentation is uncommon. However, multiple
uses for IPv6 fragmentation have been mentioned -- from NFS, to tunnels
or the recent data posted by Mark Andrews. I think such use cases should
really be considered.

That aside, if the IPv6 fragmentation function is removed, it also means
that UDP can only be used for applications that send datagrams
smaller than 1280 bytes (assuming no Path-MTUD for UDP). I haven't done
a survey myself, but I wonder to what extent one can really conclude
there's no need for that (e.g., I'm told that in the stock market sector
they employ multicast... which might mean that they need to send such
"large" UDP datagrams).


** Therefore.... **

Considering the above, I guess I'm in the camp of "avoid fragmentation
where possible". However, I don't think I'd go as far as deprecating it.

Just my two cents.

Cheers,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492