Ah! Now I see the problem. To restate it, suppose a client uses an
ISP-centric ALTO erver. It has detailed data for the ISP¹s networks, but
(as an extreme) puts everything else in one PID.
Then if the client asks for costs to (or from) the client to N servers,
all of which are in the ³other² PID, the client will get the same cost for
all N servers. So if some of those N servers have really bad connectivity,
because of conditions known only to the ISP at the remote end, the local
client will try to use those bad servers.
I don¹t think that has anything to do with upload vs download, though. It
is just a general problem that an ISP-centric ALTO server¹s costs will not
be able to reflect problems outside the ISP¹s network.
- Wendy Roome
On 03/25/2015, 18:41, "Piotr Wydrych" <[email protected]> wrote:
>Wendy,
>All,
>
>W dniu 2015-03-25 o 16:30, Wendy Roome pisze:
>> For download costs, can't the client do an ECS request with dst=my-addr
>>and src=server-addr? Or did I misunderstand your problem?
>
>Indeed, a client may ask for such a ranking (or search for needed
>information in the maps). The issue is that:
>1. If it asks a remote ALTO server, it will get the remote ISP's cost of
>uploading data from server to a client, not local ISP's cost of
>downloading them.
>2. If it asks a local ALTO server, the server may be unable to assist
>the client. I.e., it may return nothing valuable or totally nothing. To
>compute a cost of downloading the data (and/or rank endpoints according
>to the download cost) the ALTO server needs to know the download path
>(e.g., the interface through which the traffic is entering the local
>network or number of AS hops).
>In my opinion we must assume that a client connects to the local ALTO
>server, as this gives the ISP more control (and, thus, inventives to
>provide ALTO information to its clients). I can imagine easily that an
>ISP provides the service for its clients and disallows remote clients to
>use it.
>Thus, the entity that need to gather more information is the local ALTO
>server. And this is one of my primary use-cases for inter-ALTO.
>
>Best,
>Piotr 'GhosT' Wydrych
>--
>Piotr 'GhosT' Wydrych .. xmpp:wydrych//agh.edu.pl .. http://wydrych.net/
>
>A: Because it messes up the order in which people normally read text.
>Q: Why is top-posting such a bad thing?
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
