Jack Corbett (8/6/2012)
I over thought this one, and knew I was doing it, but still did it anyway. My thinking was that, while you do not have to define a clustered index as unique, SQL Server does add a uniqufier, thus I eliminated non-unique clustered index. See what I mean about over-thinking?
I was wondering why, at the time I answered, 45% of respondents didn't think a non-unique clustered index was possible. I'll give them the benefit of the doubt, and chalk it up to them all thinking as much as you. :-)
Since the clustered index is referenced by other indexes as the row identifier (since the b-tree is organized by the clustered index and therefore that would be the most efficient reference to use) a uniquifier is necessary on a "non-unique" clustered index. But since that uniquifier is not directly accessible through DML, the clustered index appears to the client as non-unique.
Hugo, I can see why you didn't want the "choose 3" text there -- for many of us it takes a question with 31 possible answers (2^5 -1 since I believe choosing 0 answers is not an option in this system down to a choice of 3 (if you got the non-unique clustered index) or 6 (if you weren't sure of the non-unique clustered index.)
But I for one am glad it was there, as I haven't done much with XML or Spatial data as yet, so I had to guess for the last one. I got it right and learned something today!