Damien Saucez wrote:

>> If i understand the proposal correctly, the additional load the Oracle
>> approach introduces is generated by the sorting of the candidates list.
>> Having this in mind, the scheme is actually very similar to the other
>> approaches, i.e. the cached lookup in the DB and socket Handling is the
>> same. Since sorting is a well known problem and can be done fairly fast
>> i don't see this as a huge overhead. But it introduces extra stress on
>> the server.
> If the sorting time becomes long, you can de activate sorting as the rank
> value is in the answer, then, the client has to rank. We should document
> this, except if the sorting of the answer is a MUST.

If i understand the draft correct both cases *must* be implemented and
supported by the server - however, the section is not 100% clear on that
- so maybe we want to state that one of the two *must* be supported
instead of both.
Currently i am not seeing these queries getting really large, so the
sorting should not put a lot of stress on the server. A BitTorrent
tracker can just download the map and do it's own ranking, while for a
single client it might be too much overhead to download the map entirely
to do just a few rankings. Since clients will not query very often this
should not be a problem.
However, the Oracle approach is dependent on the ordinal query/answer
type to be present since it does not want to give out specific weights
(in the current state anyway). It just does ranking and *must* sort the
answer to remove the original weights.

Cheers,
Ingmar

-- 
Ingmar Poese                             [email protected]
TU Berlin / Deutsche Telekom Labs
Sekr. TEL 4, FG INET                        www.net.t-labs.tu-berlin.de
Ernst-Reuter-Platz 7
10587 Berlin, Germany
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to