Le 22 avr. 2010 à 19:31, Steven Blake a écrit : > On Thu, 22 Apr 2010 10:08:34 +0200, Rémi Després <remi.desp...@free.fr> > wrote: > >> Le 21 avr. 2010 à 23:17, Brian E Carpenter a écrit : >> >>> On 2010-04-21 20:50, Rémi Després wrote: >>>> Hi Brian, >>>> >>>> I wonder what you think of what I answered to James on another >>>> discussion thread. >>> >>> I agree. I think that particular SHOULD in the RFC is an error. It >>> "SHOULD" >>> have said something like: >>> >>> "The source node MUST select new Flow Label values by a method that >>> prevents unintended Flow Label value reuse." >> >> Yes, that's more appropriate. >> >> Suggesting in addition that a 5-tuple hash can be an easy way to set >> flow-label values, because it is stateless, would IMHO improve chances > that >> host really set them. >> Wold you agree on this too? > > My reading of "The source node MUST select new Flow Label values by a > method that prevents unintended Flow Label value reuse." would preclude use > of a 5-tuple hash, which could result in coincidental selection of a flow > label value already in-use by another flow.
Well, "unintended" may be taken as permitting the hash (its intent of the hash that two different 5-tuples give in general two different values, with only statistically rare exceptions), but better words may also be proposed. In any case, explicitly permitting the 5-tuple hash is IMHO desireble. Regard, RD > > Do we really want/need to specify a MUST here? What is wrong with > low-probability, coincidental flow label reuse between flows with different > source/destination address pairs, so long as the values are otherwise > uniformly distributed and unpredictable? > > > Regards, > > // Steve -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
