Le 8 sept. 2011 à 08:18, <[email protected]> <[email protected]> a écrit : > De : Rémi Després [mailto:[email protected]] ... >> ... criteria to evaluate port-indexing methods: >> [Med] In fact we are not evaluating the algorithm at this stage but only >> characterizing them with a set of properties.
OK, just a vocabulary ambiguity. We share the same objective. >> a) Fairness (no CPE gets well-known ports 0-1023) >> [Med] We indicate whether this range is excluded or not in the properties >> table. But, this can be seen also as a loss of port utilization efficiency. Yes, it has to be considered. Now: - Giving 15 ports, as proposed in the two 4rd variants, where 16 ports can be given by relaxing the fairness constraint isn't a big deal. - After a recent exchange of information with Christian Huitema, I understand that excluding only the well-known ports 0-1023 is enough. Then the port-set-size reduction due to fairness can become 1/64 instead of 1/16, which is even more negligible. >> ... >> d) UPnP friendliness (interleaved port sets) >> [Med] I guess you are referring to IGD:1. Difficult to assess for at least >> two reasons: (1) IGD:1 is broken and (2) implementations adopt several >> behaviour (incremental or random). I-D.draft-deng-v6ops-aplusp-experiment-result is IMHO a useful reference on this subject. Having read it, I now share the view that if a port set is abundantly scattered, version 1 of UPnP will work better. This is useful because, I have been told, this version of UPnP is significantly deployed. If this is wrong, please let me know. If despite the above, the final consensus would be that IETF doesn't care about this feature, the latest 4rd address mapping can easily be simplified accordingly. Cheers, RD
_______________________________________________ Softwires mailing list [email protected] https://www.ietf.org/mailman/listinfo/softwires
