Matt Arpidone (1/30/2010)

---

Thanks Mike, I'll keep in mind that any criticism of articles or disagreement will be met with animosity. I guess my opinion isn't welcome unless I agree with the author, which, as I can clearly see now, makes total sense on a site that welcomes "peer review".

I do not know the author at all, nor have I read any of his other work, and I was very careful in my post not to resort to empty ad hominem attacks. I did express concern about the content of the article, and if you would like to disagree with me on a particular point, feel free.

Regarding the testing, I completely understand your concerns, although I find the aggravated tone of your post a little disturbing (maybe I'm reading it wrong, if so I apologize).

The reason I did not look up the unit of measure is because units are irrelevant when discussing relative speed differences. I would rather you not point out my deficiencies as a developer (of which there are many) unless they actually have relevance to my argument. Certainly, my lack of knowledge about "client statistics" doesn't discredit anything I've posted (unless I've missed something), and furthermore doesn't impair my ability to do my job at all - which makes me curious why you insist that every one of your developers know about them.

I am a little worried that I seemed harsh in my post. I had previously typed up another post (in response to the author's post after mine) about how I thought his title was misleading and that caused me to misjudge the article - along with some additional disagreements - but then I decided to just drop it. I do still stand by what I've already presented, but I recognize that I seem to have missed the true purpose of the article.

Also I find it distasteful that you point out weaknesses in my testing and indicate that you have done your own improved testing, and then you fail to post any details about what you have done. What do you mean when you say 25-30% faster when tested correctly? Faster than what? What do you mean "correctly"? Personally, I feel that my tests were adequate to get my point across. If you have numbers that disagree with mine, why don't you post them instead of giving nonconstructive retorts?

In your article you might wish to discuss the internals of how CROSS APPLY works. It might surprise you, and just might change your mind about using a UDF. I'll leave it for you to research CROSS APPLY, as I don't want to pre-empt publication of what I expect to be an excellent article with your name on it in the near future.

What did you mean here exactly? As I understand it, APPLY is similar to joining on a correlated subquery. Actually, I confess that my love of APPLY comes from it being so similar to a foreach loop in a typical programming language. I don't see what kind of crazy internal things could be going on that make it behave differently. I did try doing some research on how it works internally, and while I didn't find much helpful material, I did come across this article: http://explainextended.com/2009/07/16/inner-join-vs-cross-apply/. It explains a specific scenario where APPLY is more efficient than a query using ROW_NUMBER(), and it explains the minor pitfall with ROW_NUMBER(). I suspect something similar is happening with the product version query.

Anyway, I'm not asking you to spoonfeed me a lecture all about how APPLY works, but a link to an article describing this surprising internal behavior would be appreciated since most of the stuff I'm finding gives basic overviews and is pretty much consistent with what I expected.