This content has been marked as final. Show 2 replies
Hmm, I would guess that maybe that refers to dimensions who have many blocks queried on a continuous basis.
Have you tried flipping the dimensions around and seeing what the performance difference is? Can you even measure it? It shouldn't really be that hard -- try it one way, time it, flip it, try it again. Of course every database is different, but I would be intrigued to hear what kind of performance difference you see. Of course you'll have to control for roughly equal cache states, but it shouldn't take too long. Unless you have a 90 PAG file database in which case, yes, I can see your desire to ask before you try. I still think a medium sized of maybe 10 gigs ought to give you a good feel for it. Just my $0.02.
Hmm, I was thinking more about this and then saw John Goodwin's link in this thread:
Sort Buffer and Buffer
And that lead to reading the DBAG section on "Improving Retrieval Performance on Large Dimensions".
And that led me to wondering if you have tried your queries in MDX. I recently wrote about a really cool and undocumented (for now, I think it's coming) MDX command called NONEMPTYBLOCK which can have amazing performance given the right BSO context: