On Fri, Jul 12, 2013 at 9:51 AM, Wendy Roome <[email protected]>wrote:

> First, if discovery returns an IRD with two Network Maps, how does the
> client know which map to use? Yes, we could add metadata to describe each
> map's properties, and the client software could search that. But if we can
> do that, why not associate that metadata with the IRD -- and hence the
> server -- and have discovery do the searching?
>

Given the current discovery protocol, that effectively means the client is
going to run some algorithm to figure out which of the IRD URIs it
discovers is desired (this is assuming, of course, that the discovery
protocol is setup to return multiple URIs).  At that point, why not
discover a single IRD and then let the client figure out which one it wants
based on the metadata in each IRD?  The metadata in the IRD can be extended
in the future as needed;  the discovery protocol uses other standards
(e.g., DNS) which can be less expressive.

As for a motivation for having multiple network maps and how it can work,
consider this:
(1) Client says it is interested in the 'hopcount' metric, so it finds the
cost map that provides it.
(2) Client looks in the IRD entry for that cost map, and it says that it
depends on "fine-grained-network-map"
(3) Client finds the corresponding network map and fetches it

A different cost map can depend on a different network map.


>
> Second, I gather discovery is designed to find "the ALTO server
> recommended by your ISP". That's very useful, but that's not the only game
> it town. I think there will be plenty of ALTO servers which aren't
> discovered via that mechanism. For example, what happens if the client
> wants  the "hopcount" metric, and "the ALTO server recommended by your ISP"
> doesn't offer it?
>

This sounds like the conversation is diverging from the original one -
sounds like a separate feature request for discovery and not the base
protocol?


>
> Finally, I think the ALTO servers that participate in discovery will be
> simple "generic" servers. They'll offer the required services -- a network
> map and a cost map for "routingcost" -- but that’s it. No one writing a
> client that uses an ALTO server obtained via discovery will count on the
> server offering anything more than the required services, and hence there
> will be no reason for those generic servers to provide anything more.
>

I disagree.  I would expect other cost metrics to be made available as time
goes on.  But this also sounds like it's diverging from the original topic.


>
> - Wendy Roome
>
> From: Richard Alimi <[email protected]>
> Date: Fri, July 12, 2013 11:31
> To: Wendy Roome <[email protected]>
> Cc: alto <[email protected]>
>
> Subject: Re: [alto] Clarifying Cost Map Dependencies on a Network Map
>
> Having two separate servers would require two separate entry-points into
> the alto server discovery.  Then the client has to pick a server at
> discovery time.  This implies that ALTO server discovery has to find
> multiple servers for a client, which I'm not sure it is prepared to handle
> (and it was hard enough as it was).
>
>
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to