Hi, I’m investigating whether it would make sense to put an extra dedicated search machine in our heavily used domino-cluster.
If our search agents do full-text searches on an enormous complex database which only resides on this dedicated machine, will this somehow reduce the cpu and memory load on the regular machines?
No doubt we can get rid of a lot of individual ft indexes and reduce the indexer’s load, but what about the ft search itself (notesdatabase.ftsearch method)? Where will it really be performed?
Subject: Full-text searches on a dedicated machine
Performance end of things is kind of outside my area of FTI knowledge. I can tell you the following things will effect performance in relation to the FTI though.
-
Creating mass amounts of updates/deletions in a database that has an FTI. The full text indexer does not delete dead keys. This won’t effect lookup performance but can eat up diskspace big time. If you have agents/users doing this then a seperate search server or a fixed plan of recreating the index is the way to go.
-
Trying to search on a database that does not have an FTI will effect performance and tends not to be as good as a proper FTI.
-
If you are using FtSearch, look into lowering the maxdocs in the method and giving the user a seperate option to search on default maxdocs.
-
If a user trys a really crazy search like a large number of multiple OR / AND statements this can cause performance issues.
One example I saw was an agent doing…
[field1] = “aaa” OR [field1] = “aab” … [field1] = “azy” OR[field1] = “azz”
(not recommended to reproduce 
…
FTSearch command btw is run depending where the database is. If the database is on the server then it will run on that server. If local will run local.
You can also use views to filter down searches before you begin them. Can help with the performance.
Subject: RE: Full-text searches on a dedicated machine
Cheerz Simon, nice informative reply!
I’m doing some more tests and will report in this forum. I’m also looking at the view.ftsearch possibilities right now 