Simple vs Searched CASE statements

Reposted from Chris Webb's blog with the author's permission.

I recently had a customer show me a calculation using a CASE statement, where they had achieved a big improvement in performance when they moved from using a searched CASE to using a simple CASE. For the record, here’s an example of a simple CASE statement:

CASE [Measures].[Internet Sales Amount]
WHEN 0 THEN 0
WHEN 1 THEN 1
WHEN 2 THEN 2
WHEN 3 THEN 3
ELSE 4
END

And what the equivalent searched CASE looks like this:

CASE
WHEN [Measures].[Internet Sales Amount]=0 THEN 0
WHEN [Measures].[Internet Sales Amount]=1 THEN 1
WHEN [Measures].[Internet Sales Amount]=2 THEN 2
WHEN [Measures].[Internet Sales Amount]=3 THEN 3
ELSE 4
END

In my customer’s code they were doing lots of complex conditions involving calculated measures, but nonetheless I couldn’t work out why their searched CASE performed so much worse because the logic in both the simple and the searched CASE was the same. Therefore, in turn I asked my friends in Redmond and Marius Dumitru gave me some very useful insights that I thought I’d pass on. Basically, a searched CASE requires all previous branch conditions to have been evaluated before a given branch is chosen, which in turn makes it harder for the query optimiser to evaluate a searched CASE in bulk mode. As a result it’s always better to use a simple CASE where possible; in the worst scenarios (and I guess this is what was happening to my customer) the performance of a searched CASE can be several times worse than a simple CASE even when they seem to be doing the same thing.