Hi Brian,
Thanks for your comments. Please see responses inline.
On 10-10-22 06:21 PM, Brian E Carpenter wrote:
Hi,
This document updates Section 3 of [RFC2460] to reduce the length of
the flow label field from 20 bits to 16 bits, and in the process
creating a 4 bit reserved field.
OK, that's clear enough...
1. I do think that the justification in the draft for such a major change,
after 12 years work based on RFC 2460, is weak. Before even knowing whether
or not I like the idea, I'd expect to see much more discussion of the
rationale and the implications. It's proved remarkably difficult to widely
propagate basic changes in IP (diffserv is little used, ECN is still largely
invisible, and Re-ECN???). We all know the story on IPv4 options, especially
router alert, and we are seeing similar issues with IPv6 extension headers.
Not to mention the flow label itself.
So the chances that we actually see significant use of such reserved bits
seem to be very low.
One thing I want to clarify that this draft is not intended to exist in
isolation. If we are not changing the current definition of the flow
label, this draft does not come into play at all.
2. The downside of reducing the flow label to 16 bits is that it becomes
clearly useless as a nonce. It can still presumably provide enough
pseudo-randomness for load balancing, and it's unlikely that a single
pair of hosts will need more than 65535 unique flow labels, but 65k
is clearly too small a space to evade random-guess attacks.
Fully agree. For this reason, I think it makes sense to decide on the
intended use of the flow label before deciding on this proposal. The
ECMP use of the flow label is personally very important, but I am unsure
about the other uses. If 20 bits are required by the new targeted use of
the flow label, so be it. But if not, putting aside a few bits can be
considered.
3. By the way, we do have some spare bits, i.e. the first four bits
in every packet, where only the values 4 and 6 are used. We probably
need to reserve the value 0 anyway, and the value 5 is theoretically
needed for RFC 1819. But even so, that leaves 12 spare code points...
:-). Those bits will not be of use if we need to perform markings on
IPv6 packets (by definition Version=6).
Thanks
Suresh
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
