Mike Dougherty-384281 (12/21/2010)

---

ronmoses (12/21/2010)

---

For educational purposes, I would genuinely appreciate it if one of the folks who take issue with the "most of the time" factor could illustrate a scenario in which that script returns different results.

I read through all the comments to be sure the "order by" was already.. noted. (as if there would be anything for me to add after the first 10 posts)

I don't have time to make a reproducible illustration but wanted to throw in these two cents: With large data sets and parallelization you can get apparently unordered results even with the clustered index in effect due to the merge step simply combining multiple streams. Without an explicit order by clause, you get the unsorted merged streams. I discovered this trying to remove the cost of the order by under the erroneous assumption that the clustered index made the order by unnecessary. No, it's very necessary.