On Aug 14, 2013, at 1:08 PM, Fernando Gont wrote: > On 08/14/2013 04:29 PM, Brian E Carpenter wrote: >>> >>> That's my point: is that a sensible requirement? For instance, for the >>> time being (Flow Label not used), there's no other way to identify flows >>> than to break this reuirement. >> >> Which is indeed why I worked hard to get RFC 6437 written. But changing >> this aspect of IPv6 design would be a *really* big discussion. > > I'm fully in favor of RFC6437. My point is that host implementations > still not use or o weird stuff with the Flow Label -- hence it's not > dependable for anything. This means that in the short of near term > there's no better option than following the IPv6 header chain (not that > I like it but... that doesn't change the facts).
As an implementor, I'd far rather hear, "Get on the ball implementing flowlabels, and until you do, any packets you originate might suffer re-ordering via ECMP because we can't reliably get to the L4 info," rather than, "Tell your customers that they can't send fragments over IPv6 (or use any other IPv6 options)." Kevin kevin.la...@oracle.com -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
