Wendy,
I am not so sure you talked yourself into and then out of the idea of data
mining as you proposed.
As for logging misses in ECS I would say that the Server (really an
aggregator in this case) will query its source of information in order to
ascertain the source. In this case the Client can never do this *unless*
the servers permitted it. This allows for Not Authorized to be returned as
a valid code or we could opt for a "origin" property to return 3 values
('self to denote it is the origin of the data, IRD of *its* source which
does not imply it is the source of information or something polite such as
'undefined' to hide a number of cases whether the server declines to
respond). Other values such as 'declined' could be more appropriate
(actually I may have too much fun defining different return values and will
stop here...)
As for the popularity you may have a point. Without aggregation the
servers would have been directly queried by Clients and could see the level
of popularity but it has the down side that Clients must do all of the work
themselves as well.
As for an operator of ALTO data mining the queries I would totally see
that. It may not be something we need to standardize but it is a great
idea and would make for a prudent product feature imo.
Lyle
On Thu, Oct 29, 2015 at 12:28 PM, Wendy Roome <[email protected]>
wrote:
> In the interim meeting, Lyle mentioned logging ECS misses, and try to get
> missing cost points.
>
> How about taking that one step further, and log popular sources &
> destinations, even if the server has cost data for them? E.g., suppose N
> different clients request costs from endpoint X. That implies X is a
> popular server, and the ALTO server should try to get cost data from X to
> as many other endpoints as possible.
>
> Which leads to data mining ECS queries. Eg, collect ({src-eps}, {dst-eps},
> client-ep) tuples. From this we could find popular sources & destinations,
> and detect clusters of ALTO-using clients.
>
> Of course, that does raise security issues. Eg, someone could use that to
> discover that computers in the office of (insert politician name here) do
> ECS queries for known X-rated servers. This might discourage clients from
> using ECS, and encourage them to use full cost maps instead.
>
> - Wendy Roome
>
>
> _______________________________________________
> alto mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/alto
>
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
