> Thanks, Paul. You nailed it. At Fourthought, we do practical things with
> RDF. Period. I personally am mostly curious about the Semantic Web, but it
> is not what impels me to use RDF. We have used RDF technologies and 4Suite to
> implement projects for clients with remarkable efficiency. I can disclose
> that one of these clients, the Global Knowledge Engineering division of Sun
> Microsystems, ended up with a very practical and valuable extensible metadata
> repository based on RDF, including viewing and management applications. One
> senior observer of the project noted that he had seen another such project in
> another company take ten-fold of the time and ten-thousand-fold of the budget
> to create the same thing using proprietary technology. This is no
> exaggeration.
It occurred to me that the picture I posted in my message might have come off as too rosy. Let me detail a few of the downsides of what we do with RDF and 4Suite in practice, again looking as closely as NDA allows at the Sun project.
* Maturity: 4Suite really only coalesced the needed XML and RDF processing features in its 0.12.0 version. Unfortunately, this version is still in alpha state, and noticeably so. Sun was successful with it because they had me, one of the core developers, on contract. I admit that the 4Suite project (which is open source, and happy for any help) has a ways to go before others can easily attain such success out of the box. Also, 4Suite is in active development and 0.12.0 is inevitably a moving target.
* Performance: The initial RDF model for sun involved between 300,000 and 500,000 RDF statements. Query and update performance has been a problem, which we've addressed by caching, and with some optimizations to the DBMS and Versa query language. I think we've made impressive progress on this, but against SQL-like benchmarks, we have a ways to go. Ken North put forth a nice canned expectation of current technology on commodity hardware: 1.15 seconds to return 50,000 rows for SQL (I assume with at least one join) and 2-3000 searched docs per second for XQuery. I would informally put current 4Suite RDF DBMS performance on commodity hardware in the 10 seconds per 50,000 retrieved resources (including at least one edge traversal). This was probably closer to 500 seconds when testing began in the Sun project in March, and as much as 100 seconds just a couple of weeks ago. (Ain't refactoring grand?). This is actually close to my goals, which I think will need comp-sci heavy!
weights to push the remaining order of magnitude to SQL or XQuery scale.
* Technological stability: it is not by any means certain that RDF will continue along its (IMO) successful path. Some of the technologies we use, e.g. RDFS and DAML+OIL are certain to change regardless of RDF's success. We are currently working on a risk analysis plan to make sure the project can remain useful in the face of such changes. However, contemplating change in the foundation of the technology on which you build an application gives any conscientious project manager cold night sweats.
--
Uche Ogbuji Principal Consultant
+1 720 320 2046
Fourthought, Inc. http://Fourthought.com
4735 East Walnut St, Boulder, CO 80301-2537, USA
XML strategy, XML tools (http://4Suite.org), knowledge management
