Re: Linked Data Fragments: Web-scale querying

Hi Martynas,

I'm fully with Olaf here, except for two side-remarks:

>> How does LDF compare to [SQUIN]?
> 
> I understand LDF more as an alternative practice for publishing (Linked) data 
> on the Web. However, the primary purpose for proposing this practice (as Ruben 
> and his colleagues emphasize in their LDOW paper) his to facilitate 
> distributed querying without putting most of the burden on the servers.

So distributed querying is indeed an important part of what we want to achieve.
In general:

SQUIN: solving queries through Linked Data dereferencing
Linked Data Fragments: publishing Linked Data in a low-cost, queryable way
Linked Data Fragments: client: solving queries through Linked Data Fragments

> So, the primary difference between querying LDFs and the aforementioned Linked 
> Data query processing is that querying LDFs is a form of distributed query 
> processing, that is, some part of the execution of a given user query is 
> distributed to LDF servers (which provide a limited form of query processing 
> functionality). In contrast, Linked Data query processing  is _not_ a form of 
> distributed query processing. Instead, for Linked Data query processing 
> approaches, the whole execution of a query happens within the Linked Data 
> query processing system (e.g., SQUIN), simply because these approaches do not 
> assume / rely on server-side query processing functionality (servers in this 
> setting are only required to answer URI lookup requests).

Here, it depends on how you define "querying".
You could say that a regular Linked Data server offers querying,
because you give it a URI, and it will look up the resource associated with that URI.
So in that sense, regular Linked Data servers offer a { <s> ?p ?o } querying interface.
For instance, few would disagree that
    http://dbpedia.org/page/Barack_Obama
is the result of querying.

For another definition, you could say that basic Linked Data Fragment servers
_don't_ do querying: they just offer a finite set of resources you can retrieve.
Only it's not dereferencing, but a little more flexible.
Or you could say they do offer { ?s ?p ?o } querying; depends on how you look at it.
But note that basic LDF servers do not necessarily need a query processor;
they might as well offer a set of pre-generated HTML files.

So the main difference with regular Linked Data servers
is that basic Linked Data servers offer a more specific interface
that just HTTP dereferencing;
each basic Linked Data Fragment is subject to these constraints:
- they offer data corresponding to a triple pattern
- they offer the total count of all patterns that match the triple pattern
- they offer hypermedia controls to all basic LDFs of the same dataset

Best,

Ruben

Received on Wednesday, 19 March 2014 15:29:18 UTC