Hi Bill, Ben, Dhody and Rich
An ALTO Server could definitely consider cases A and B, while
providing rules in case the values are provided in Numerical mode.
In case A, Bill's rule can be adapted to 2 cases to be specified:
1 - If the numerical value represents a cost (e.g. routingcost) to be
minimized then Bill's rule applies
2 - If the numerical value represents a utility (e.g. available
bandwidth) to be maximized, then this rule becomes:
num(A,B) >= num(C,D) implies ord(A,B) <= ord(C,D),
and so on...
In case case B, the rank is typically provided according to knowledge of ALTO server only (policy or other qualitative consideration not to be unveiled to clients).
In that case, I guess no numerical value is provided, no compaptibility rule is needed.
Sabine
Ben Niven-Jenkins a écrit :
I guess it depends on whether the semantics of ordinal costs is:
A) rank according to numerical costs; or
B) rank according to something the ALTO server knows
If you consider a future case where people define additional cost types the (A) requires them to also define both an ordinal and numeric cost type (assuming both are meaningful) whereas (B) doesn't and gives the ALTO server freedom to do what it likes.
I don't have a strong opinion either way.
Ben
On 3 Oct 2012, at 18:07, Richard Alimi <[email protected]> wrote:
I can't think of a valid reason why they should be inconsistent. I'd
be fine with mandating it in the protocol, but I'd have to leave that
up to the chairs (and of course, feedback from the rest of the WG)
given that it has already been through WGLC and has been submitted to
IESG.
Rich
On Wed, Oct 3, 2012 at 9:23 AM, Bill Roome <[email protected]> wrote:
Suppose an ALTO server supports both "numerical" and "ordinal" cost modes
for the same cost type. Should the protocol require that the values for
those two modes to be mutually consistent?
And if we do want that, how should the requirement be worded? I'm tempted
to say
"Let num(A,B) and ord(A,B) be the numerical and ordinal costs
for source PID A and destination PID B, respectively.
Then for all PIDs A, B, C and D,
num(A,B) >= num(C,D) implies ord(A,B) >= ord(C,D),
ord(A,B) >= ord(C,D) implies num(A,B) >= num(C,D),
num(A,B) <= num(C,D) implies ord(A,B) <= ord(C,D),
ord(A,B) <= ord(C,D) implies num(A,B) <= num(C,D)."
Is that sufficient?
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
--
---------------------------------------------------------
Sabine RANDRIAMASY
Alcatel-Lucent Bell Labs France
Centre de Villarceau
Route de Villejust - 91620 NOZAY - FRANCE
E-MAIL : [email protected]
TEL: +33 (0)1 30 77 27 45
(On Net) 2 103 27 45
---------------------------------------------------------
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
