I read a post about searching from Bex Huff last week and I immediately felt that his post was worth sharing. If you haven’t read it yet, take a moment to do so now. I think he makes some good points and I want to build and not just repeat.
I’ve been on some projects recently that have had to depend on full-text searching. When I joined one project, search results were extremely slow. While we re-architected the solution, we limited the number results returned to enhance the speed. Once we fixed the search, we debated what we wanted the new limit to be for the users.
By default, Documentum limits results to 350 items per search. This can readily be changed with a client/web server setting. However, we discussed using a different number when the time came to raise the limit from the temporarily low number. It was one thing when we were on system defaults to not document the limit. Now that we had published one lower limit in the system documentation, we couldn’t just say that there was no limit.
Bex suggests a number of 200. If you have more than that, refine the search. In our discussion, we determined that in some circumstances, 100 would be too many, and in other circumstances, 500 might be realistic. 500 you say? We are dealing with images, and sometimes in our scenarios, the scanning order is the determining factor and the ability to quickly scan through a hit list is useful.
In the end, we punted and went back to the default 350 and just documented it. Now we will learn if our Information Architecture is as good as we think it is.
3 thoughts on “Control Those Search Results”
In my experience the biggest problem with searching a doc object repository is when there are too many sub-doc types. I have seen doc objects with 3-5 layers – and the search engine grinds to a crawl.
Alan, that is a problem and goes back to a good design as that can impact performance across the board. A slow search that returns a lot of rows would kill any system quickly.
500 items for an image search makes sense… some of my Flickr sets have over 1000, and its not too tough to scan through them quickly to find what I want.
Comments are closed.