Re: Linked Data Fragments: Web-scale querying

great! I have tried in the past the construction of  a sort of "federated"
Linked Data "cache" server, and from that experience I was looking for some
library who decompose queries in order to maximise triples re-using. If I
am not wrong is what you are doing here?
Most of of all I am curious about the methods: do you propose this approach
as a pattern more than a technologic solution?

If I undertand correctly the idea, this seems to me great, and can be
expanded in many ways, maybe also introducing a sort of "linked hypermedia"
layer.

I'll follow the updates! :-)







2014-03-11 13:06 GMT+01:00 Ruben Verborgh <ruben.verborgh@ugent.be>:

> Dear Semantic Web and Linked Data enthusiasts,
>
> If you're curious about new ways to query Linked Data,
> you might like our Linked Data Fragments client.
> It lets your browser execute SPARQL queries over Web data
> in a scalable way: http://client.linkeddatafragments.org/
>
> Today's answer to Web querying consists of SPARQL endpoints.
> Publishers of data sets offer a public endpoint, which
> answers highly specific questions for any client. Unfortunately,
> the availability of public SPARQL endpoints is problematic [1] –
> and thus so is the availability of publicly queryable datasets.
> We cannot rely on them for building applications, and that's a pity.
>
> It is not an issue of performance but an inherent architectural issue:
> making a public server responsible for arbitrarily complex requests
> doesn't work on a Web scale. We have to create more simple servers,
> only answering simple questions that don't endanger availability.
> Yet at the same time, the dataset should remain easily queryable.
>
> This is the goal of the Linked Data Fragments project [2].
> “Linked Data Fragments” is a term for all ways to offer parts of a dataset:
>  - SPARQL results are (precise but expensive) Linked Data Fragments.
>  - Data dumps are (huge but straightforward) Linked Data Fragments.
> Between those two extremes, an underexplored range of fragments exists.
>
> We propose a new type called “basic Linked Data Fragments”,
> which partitions a dataset in all its basic triple patterns.
> This reconciles the need for queryable public datasets
> with the availability demands of Web applications.
> A basic Linked Data Fragments server with well-known datasets
> is available online [3] (and so is its source code [4]).
>
> Try our online client [5] that answers SPARQL queries
> using only basic Linked Data Fragments (source code [6]).
> It works up to two magnitudes faster than Linked Data Querying [7]
> because servers offers those fragments that assist client-side querying –
> without needing to solve expensive queries at the server side.
>
> Basic Linked Data Fragments are not a definitive answer;
> there are many other types of fragments to explore.
> However, you might be surprised to see quite acceptable query times,
> and – most importantly – high availability and scalability.
>
> Read more on Linked Data Fragments on the website
> http://linkeddatafragments.org/ and discover all details
> in our forthcoming LDOW2014 publication [8].
>
> Looking forward to your feedback!
>
> Best regards,
>
> Ruben Verborgh
> Ghent University – iMinds, Belgium
>
>
> [1] http://sw.deri.org/~aidanh/docs/epmonitorISWC.pdf
> [2] http://linkeddatafragments.org/
> [3] http://data.linkeddatafragments.org/
> [4] https://github.com/LinkedDataFragments/Server
> [5] http://client.linkeddatafragments.org/
> [6] https://github.com/LinkedDataFragments/Client
> [7]
> https://cs.uwaterloo.ca/~ohartig/files/Hartig_LDQueryExec_DBSpektrum2013_Preprint.pdf
> [8] http://linkeddatafragments.org/publications/ldow2014.pdf
>
>