On Fri, May 17, 2013 at 9:19 AM, Wendy Roome <[email protected]>wrote:
> I'd make an even stronger statement: a client MUST NOT compare ordinal > costs between different response messages, even from the same service. This allows better better concurrency and should be clarified. > A > client MAY compare one ordinal value against another ordinal in the same > response, but that's all. Why add a MAY? The current wording (Sec. 6.2) is "If the Cost Mode is 'ordinal', the Path Cost of each communicating pair is relative to the m*n entries." In other words, a response returns m*n entries, where each entry is a number (ranking). The current spec does not specify that the m*n numbers are distinct, and hence the response may contain entries with the same value. Consider an example of two two sources (S1, S2) and three destinations (D1, D2, D3), and a response as the following: S1 -> D1: 1 S1 -> D2: 2 S1 -> D3: 2 S2 -> D1: 1 S2 -> D2: 2 S2 -> D3: 3 The interpretation is (~ means equivalent): S1 -> D1 ~ S2 -> D1 < S1 -> D2 ~ S1 -> D3 ~ S2 -> D2 < S2 -> D3 In other words, it defines a partial ordering. Is this what you are referring to in terms of MAY? Richard > For example, suppose a client fetches a full ordinal cost map twice in > close succession. The ordinal values in the second response may be > completely different from those in the first response, even if there was > no change in underlying network performance. > > Put it another way: When a client requests an ordinal cost mode, the ALTO > Server MAY apply a non-linear order-preserving transformation to the > underlying numerical costs, and that transform MAY vary from request to > request. > > IHMO, I thought the reason for defining ordinal costs was to allow an ALTO > server to obfuscate the underlying numerical costs. So if the server was > really paranoid, and will to expend the cycles, it might very well use a > different obfuscation algorithm for each request. > > - Wendy Roome > > > >Date: Thu, 16 May 2013 21:51:30 +0100 > >From: Ben Niven-Jenkins <[email protected]> > >Subject: Re: [alto] Equivalence of various identifiers from an ALTO > > Server > > > >(BTW I'm not sure you can compare a filtered costmap against a full cost > >map if the costs are ordinal) > > > >Ben > > > _______________________________________________ > alto mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/alto >
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
