- From: Lee Feigenbaum <lee@thefigtrees.net>
- Date: Fri, 12 Mar 2010 09:41:46 -0500
- To: Ivan Mikhailov <imikhailov@openlinksw.com>
- CC: RDB2RDF Working Group <public-rdb2rdf-wg@w3.org>
On 3/12/2010 5:33 AM, Ivan Mikhailov wrote: > Hello Lee, Hi Ivan, Thanks for the response! Comments below... >> Cambridge Semantics plans to implement the RDB2RDF standards in our Anzo >> software... we've identified 2.5 requirements for the >> RDB2RDF mapping language: >> >> 1/ Tooling >> >> It must be straightforward to create tooling that generates mappings. >> Most of the candidate designs I've seen have no problem here, but I >> wanted to include it anyway. Declarative XML- or RDF-based mappings are >> very easy to target. Mappings based on less structured query forms (SQL >> views or SPARQL CONSTRUCT queries) are less desirable but are doable. >> From our perspective with respect to tooling, a SPARQL CONSTRUCT-based >> approach is are preferable to a SQL DDL-based approach. > > CONSTRUCT approach does not provide enough information about mapping to > SPARQL-to-SQL optimizing compiler. Efficient mapping requires much more > details than CONSTRUCT may fit. It would be nice to use CONSTRUCT but > we've made much more complicated language to not kill the performance. Can you talk more about the requirements that you've identified for a mapping language for performance reasons that go beyond the expressivity of CONSTRUCT queries? Is there a write-up of this somewhere? I don't have implementation experience here myself, but would definitely like to hear Eric P's take on this, as well. >> 2/ Named Graphs >> >> The mapping language should be capable of mapping a relational schema >> into SPARQL's named graph model. The granularity of the graphs should be >> tweakable within the mapping -- for example, we sometimes will map an >> entire table into a single graph, and other times we will map specific >> rows/resources into their own graphs. > > We've found that mapping of graph does not differ much from mapping of > other items of a quad. So I absolutely support this requirement. > > > >> ...and a half/ Update >> >> I realize that updating the relational database is beyond the scope of >> our charter. That said, we would like to consider the potential >> extensibility of the mapping approach chosen to handle (some cases of) >> writing data back to a relational schema. (Either via SPARQL Update or >> via triple/quad removes& adds. >> >> (As we begin to compare and contrast design candidates, these are some >> of the criteria that we'd like to use to evaluate the possibilities.) > > I've written http://esw.w3.org/topic/UpdatingRelationalDataViaSPARUL > as a seed for a discussion but it did not cause any reaction so I > postponed the implementation. I'd like to hear any comments/wishes, even > strong reject is better than nothing. Excellent & thanks - I'll take a look at that soon and send comments. Lee > > Best Regards, > > Ivan Mikhailov > OpenLink Software > http://virtuoso.openlinksw.com > > P.S. This is my personal opinion, not a formal reply from any W3C > working group. > > >
Received on Friday, 12 March 2010 14:42:50 UTC