Version 1 (modified by joshuadf, 10 years ago) (diff)


Meeting minutes from 11/27/2007


Discussion was mainly about the merits of different approaches to materializing large queries, such as full transitive closure of the entire FMA.

Marianne presented some of her thoughts from attempts to "explode" transitive closure at the database level.

She first with a postgreSQL stored procedure which took approximately 8 hours but resulted in a 100GB+ database size.

Marianne also experimented with SQL99 recursive queries in MS SQL Server. Unfortunately computing transitive closure over the entire FMA took days (we theorized that the DBMS was running out of memory and so was frequently forced to swap to disk).

Neither approach currently provide any performance advantages vs Todd's transitive closure implemented as high-level Java code (an extension to Jena). However, Dan Suciu pointed out that it is guaranteed to fail in the general case (for example, a query like "Give me the FMA plus some other nodes") due to memory constraints, while a DBMS will at least swap to disk. Additionally, we need to support large queries for the practical reason of providing fully materialized ontologies to third parties.