On 05/18/2012 15:00, "[email protected]" <[email protected]> wrote:
>On Mon, May 14, 2012 at 2:24 AM, Songhaibin <[email protected]> >wrote: >>Hi Rich, >> >> >> >>>-----Original Message----- >>>From: [email protected] [mailto:[email protected]] On >>>Behalf Of >>>Richard Alimi >>>Sent: Sunday, May 13, 2012 2:05 PM >>>To: Songhaibin >>>Cc: Enrico Marocco; [email protected] >>>Subject: Re: [alto] WGLC: draft-ietf-alto-protocol-11 >>> >>>On Mon, Apr 23, 2012 at 6:47 PM, Songhaibin <[email protected]> >>>wrote: >>>> In general, this draft is in good shape. >>>> >>>> One randomly caught problem, in Section 5.1.2.2, the ordinal mode >>>introduction, it says this cost mode is a ranking from a particular >>>source to >a set of >>>destinations. But in Section 5.2, the introduction of cost map >>>structure, the >cost >>>mode "ordinal" can be applied to m*n entries in the cost map. So I >>>suggest to >>>modify the previous section 5.1.2.2 to make them consistent. >>>> >>> >>>Thanks for catching that. In fact, the sentence "If the Cost Mode is >>>'ordinal', the Path Cost of each communicating pair is relative to the >>>m*n entries" can probably just be removed. >>> >> >>I'm sorry that I have a different opinion here. > >No need to apologize :) > >[Haibin] I only pretended to be. :) > >>The "ordinal" cost to entries in a matrix makes sense to me. I prefer to >modify the definition in 5.1.2.2, the ranking can be from multiple >sources to >multiple destinations. Or is there special reason to limit it to 1*n? >> > >As an author/editor, I will say that the intent was for them to be >relative to only the costs from the same source. However, if the WG >believes that should not be he case, then please do say so. > >To summarize the differences in semantics: >(a) The interpretation that an ordinal cost is relative to all other >entries in the cost map means a server could convey a policy like "I >would rather you send traffic from s1 -> d1 instead of from s2 -> d2" >(b) The interpretation that an ordinal cost is relative only to other >ordinal costs from that same source means you can't specify that >policy in an ordinal map. > >The original thought was that (b) makes the protocol a bit simpler to >reason about and captures the major use cases for ordinal costs. >However, it has slightly less expressive power when it comes to >encoding policies. It is certainly worth having the discussion as to >which way we should go. > >Any other comments? > >[Haibin] I can give a use case where "option (a)" is meaningful. I think >other people might give other use cases too. Assume that in CDN >interconnection scenario, there are several surrogates from downstream CDN >that can be injected delegate contents from upstream CDN, and at the same >time there are a few surrogates in the upstream CDN that can provide the >content. It is better to query the ALTO server do decide a path from a >particuplar surrogate of upstream CDN to a surrogate of downstream CDN. >The >ordinal costs applied to a m*n entries can be used for this decision. I always assumed "option a": ordinals are relative to all costs in this response, not just the costs for the same source. Of course, "option a" automatically satisfies "option b", so if a server is willing to do (a), then the server is in compliance regardless of what we decide. - Bill Roome _______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
