2011-06-11Myth: Most Selective First

Every time a compound index is created, the order of the columns must be chosen wisely. Section 1.2 is devoted to this question.

However, there is the myth that you should always put the most selective column to the first position; that is just wrong.

After that, there are even reasons to put the least selective column first. The Oracle database can, for example, use an INDEX SKIP SCAN in that case. But that’s an advanced feature. The most important factor...uhm, did I say that before?

The true essence of the myth is related to indexing independent range conditions—that is the only case where the selectivity should influence the index design (see Section 4.3).

The myth is extraordinarily persistent in the SQL Server environment and appears even in the official documentation. The reason is that SQL Server keeps a histogram for the first index column only. But that means that the recommendation should read like "uneven distributed columns first" because histograms are not very useful for evenly distributed columns anyway.

I’m not the first to fight this myth. Here are some more references that disproof the myth: