Paul White (3/13/2010)
It does raise a question, though: in which order should we allocate?You have chosen 'cars', then 'trucks', then 'RVs' - but that seems a bit arbitrary.
I initially thought of that, too. But I believe that there will be a nice balance to the 3 insofar as which will have the largest numbers... the first random selection has just as much a chance of being very low as it does very high. Same goes for the others. During some very minor testing, for example, it did produce 3,3,16 a couple of times. I've not done an actual "distribution" test yet to see if it's more random than not, but mathematically, it would seem so. Guess I'll have to do a test because, heh... one test is worth a thousand "expert opinions" (especially mine :-P).
My second point concerns performance.
The code performs three partial scans of the index, and three sorts.
CPU time on my machine: 188ms - averaged over ten runs.
Combining the two approaches reduces the average time to 94ms:
I was thinking the same thing afterwards. Considering that I just did a review on your Cross Apply article, the notion struck me that your method and mine could be combined for the best of both worlds, but you beat me to it. You're much better at the use of Cross Apply and it seems to just come natural while I still have to think about it. :blush: It'll become more natural for me as I use it more, but you're still light years ahead of me on its usage.
--Jeff Moden
Change is inevitable... Change for the better is not.