Re: Enquire - WWW - Semantic Desktop/PersonalDataWiki? do you agree?

In looking at the benchmarks, the one that was a mapping of SPARQL onto a relational database was just about as fast as the examples 
that used the relational databases directly.  This indicates that it isn't the query language or the data model that fundamentally 
caused the performance problem, but the storage model and implementations that are broken efficiencywise.  Triplestores are 
extremely flexible which comes with some cost, however I feel that they can be improved with some dynamic and semi-static 
optimization to approach RDBMS efficiency.  That efficiency is separate from whether RDF / semantic databases are a better model for 
architecture and development.

sdw

On 11/23/10 2:46 AM, adasal wrote:
> Hi Stephen
> I will look at the reference and come back to you. I understand you to
> be indicating that the triple store back end in the Berlin benchmarks
> is a naïve implementation. Have I understood? If so could you
> elaborate? Are there OS alternatives? Without reading through again I
> thought it was Jena backed by Postgres. I thought Jena was acting as
> an ORM autocreating the table schema. As I say I don't have info in
> front of me. Every bit of knowledge here is a valuable nugget at the
> moment. Thanks. Adam
>
> On 22/11/2010, Stephen Williams<sdw@lig.net>  wrote:
>> On 11/22/10 2:01 PM, adasal wrote:
>>> On 22 November 2010 18:19, Stephen Williams<sdw@lig.net
>>> <mailto:sdw@lig.net>>  wrote:
>>>
>>>      Getting the enabling technology and paradigms right is a prerequisite
>>> for such a solution, but they are not sufficient.
>>>       Jumping beyond our current plateau is going to take more than the
>>> simple application veneer that were enough for most
>>>      generations of solutions.  Too many people are constrained to thinking
>>> in the language of existing software elements.  This
>>>      shows even with techies by the slow adoption of triplestores / SPARQL
>>> / et al vs. RDBMS systems, which are clearly deficient.
>>>
>>>
>>> They are not in the least obviously deficient. They are extremely
>>> efficient and timely in what they deliver. Triple store returns
>>> results in slow motion compared to RDBS. Look at the Berlin benchmarks.
>> Not all applications are performance intensive.  Sometimes, the worst
>> bottlenecks are development, evolution, flexibility, and
>> maintenance.  RDBMS is a poor model for many things, with current
>> applications force fitting models into it.  In the March Berlin
>> benchmark [1], if you look at the D2R SPARQL->RDBMS mapping, you can see one
>> solution to having an efficient SPARQL database.  It's
>> not the model that is bad, it is the naive implementation.  We are certainly
>> going to find multiple solutions to data clustering
>> that will provide high performance.  Another thing to keep in mind is that
>> the market moved from hierarchical databases to
>> relational even though there was a performance loss.
>>
>>> Notice what project inspired those benchmarks.
>>> Where is this plateau I am jumping beyond? I'm not sure I have reached it
>>> yet.
>>> This is the language of a convinced evangelist. But it isn't a convincing
>>> rational argument.
>>>
>>> Adam
>> [1]
>> http://www4.wiwiss.fu-berlin.de/bizer/BerlinSPARQLBenchmark/results/index.html
>>
>> Stephen
>> --
>> Stephen D. Williams sdw@lig.net stephendwilliams@gmail.com LinkedIn:
>> http://sdw.st/in V:650-450-UNIX (8649) V:866.SDW.UNIX
>> V:703.371.9362 F:703.995.0407 AIM:sdw Skype:StephenDWilliams Yahoo:sdwlignet
>> Resume: http://sdw.st/gres Personal: http://sdw.st
>> facebook.com/sdwlig twitter.com/scienteer
>>


-- 
Stephen D. Williams sdw@lig.net stephendwilliams@gmail.com LinkedIn: http://sdw.st/in V:650-450-UNIX (8649) V:866.SDW.UNIX 
V:703.371.9362 F:703.995.0407 AIM:sdw Skype:StephenDWilliams Yahoo:sdwlignet Resume: http://sdw.st/gres Personal: http://sdw.st 
facebook.com/sdwlig twitter.com/scienteer

Received on Thursday, 25 November 2010 22:55:53 UTC