Re: bootstrapping decentralized sparql

> Sandro Hawke wrote:
> > I agree it'd be nice, but I don't see how the market/social forces could
> > be set up to make that work.

I posted this, and then an hour later, while driving, it occured to me
how to do it....

> > How would you decide which trackers would be authoritative for which
> > keywords or which data sources?  I don't see a good way to divide the
> > problem space and make the system scale, with the right incentives on
> > the right players.
> 
> if a tracker already provides an index on classes and properties,
> a keyword index is a small overhead (especially with highly
> optimised keyword indices such as Lucene).  The trick is to
> find the sweet spot between indexing overhead and benefit
> I guess.
> 
> Deciding which keyword query to send to which tracker is
> tricky.  If trackers link to other trackers, keyword queries
> could be routed along these links, but that wouldn't
> guarantee web-wide completeness.
> 
> For smaller communities of users that maintain networks of
> trackers the approach might be reasonably complete though.

I originally understood you to be asking for a way to search for a
pattern like this:

       { ?subject ?predicate "Sandro" }

(with some added keyword functionality that I'll ignore for now).
That's a query I don't know how to decentralize.  But in most cases, I
think the query will really be like this:

       { ?subject foaf:firstName "Sandro" }

and that a search can be coordinated by the tracker for foaf:firstName.
It can chose to index literals if it wants to.  The system works either
way, but the performance may vary dramatically.  Fundamentally, this
decision can be made by the vocabulary publisher, in chosing their
backlink tracking provider.  If they want folks to have faster searching
(and perhaps pay more for the tracking service), they'll arrange for a
tracker which does this kind of indexing.

I think that design will work, with the right incentives all around.

(Or, close to it.  I still don't quite know what will motivate ontology
publishers....   But there are lots of possibilities.)

     -- Sandro

Received on Wednesday, 20 May 2009 01:31:15 UTC