I found this article on Slashdot. It makes an important point that I could have benefited from during my early years at Pangea when we were building the GeneThesaurus genomic data warehouse.
Following the conventions of the day we built the system in Oracle using an "object network model" schema whose complex link types made the project feel like the "Tower of Babel". The performance and usability was a disaster as the Biologist users had expectations that their simple keyword searches should be performed on all fields.
A second iteration implemented a "Word" table leveraging bitmap indexes. This too failed due to the length of time it took to build the indexes after each "nightly" update. After camping out at customer sites for several weeks to hand-hold the updates and re-indexing process this too was abandoned.
The last iteration was the simplest but also the most successful. Leveraging just 2 software engineers we reproduced all the useful functionality using a free text search engine (AOL's PLS). The text search repository was also much easier to deploy.
