jump to navigation

Outlier Values – An Enemy Of The Index December 13, 2007

Posted by Richard Foote in Index Access Path, Indexing Tricks, Oracle Cost Based Optimizer, Oracle General, Oracle Indexes, Outlier Values.
10 comments

Outlier values are basically values that sit way way outside the standard range of a column’s normal value range.

Data can be a funny thing and sometimes there are values that are naturally “exceptional”. However, very commonly, outlier values are used by applications to represent bizarre default values, to avoid confusion with legitimate values. For example, I look after an application that uses the American Date Of Independence as it’s “default” date.

Usually, these weird outlier values are used to avoid nulls values, as nulls can be problematic and can not be indexed (well actually you can index a null column but we’ll leave that for another blog entry).

However, outlier values while (maybe) solving one problem, can introduce some very significant problems in return.

Firstly, the CBO “hates” outlier values as it potentially totally screws up the CBO’s selectivity calculations. The selectivity of a range scan is basically calculated by the CBO to be the number of values in the range of interest divided by the full range of possible values (IE. the max value minus the min value). Therefore if this calculation is invalidated by a massive and disprotionate “hole” in the full range of possible values, the CBO can get things horribly wrong.

See here for a simple demonstration:  Outlier Selectivity Problem

Additionally, indexes “hate” outlier values as it prevents Oracle using the 90-10 block split to keep indexes nice and compact and is forced to use 50-50 block splits instead. Basically a 90-10 block split is considered if and only if the index entry to be inserted is equal or greater than the current maximum value.  An outlier value that is also the maximum value,  usually means monotonically increasing values (such as sequences, dates, etc.) don’t actually insert the maximum value. Therefore, not only do indexes perform 50-50 splits but this 50% of free space is never used, as all new values are all almost, but not quite, maximum values.

Little demo to highlight this problem: Outlier Index Space Utilisation Problem 

In summary, avoid outlier values if at all possible.  They generally cause more problems than they solve !!

Follow

Get every new post delivered to your Inbox.

Join 1,861 other followers