IBM FileNet P8, Version 5.2.1            

Dynamic switch search does not choose database first when the expected hits are more than the threshold

In a dynamic switch search, the database first algorithm is not chosen by the IBM® Content Search Services server when the expected number of hits are more than the threshold value.

Symptoms

When you run a dynamic switch search, the database first algorithm is not chosen by the IBM Content Search Services server.

Causes

If the number of hits in the IBM Content Search Services index is not much more than the threshold, the dynamic count estimate can be less than the threshold because the dynamic count estimate is only an approximation of the count. The dynamic count estimate works best for protecting against a long running CBR first search when there are at least 200K items in the collection and the CBR hit count exceeds the threshold by a sufficient amount.

For example, with a single collection and a threshold of 10,000, if there are 20,000 or more CBR hits, dynamic switching tends to choose database first. But if there are only 15,000 CBR hits, there might be some searches that remain as CBR first searches because of the possible inaccuracy of the estimate.

If there is more than one index, the estimation algorithm assumes that the CBR hits occur uniformly across indexes, such as a search on a common word that can occur in many documents. Because the estimation algorithm searches the largest indexes first, if there are few hits in the initial indexes or any large indexes, the server estimation process might time out (the default is 30 seconds) or be terminated by the Content Platform Engine server. To ensure a quick dynamic count estimation time, counting will not continue past 12 collections if the CBR hits are not occurring uniformly and have a sufficient likelihood of being found across all collections.

Resolving the problem

Contact IBM Software Support for possible solutions to override the default behavior or force a database first search by using the SQL option CBR_DB_FIRST. For information, see CBR Query Optimization. If there is more than one index, ensure that the threshold is at least as large as the recommended size in the Best Practices table: Best Practices.


Last updated: October 2015
p8pcs022.htm

© Copyright IBM Corporation 2015.