>There's an argument against newsequentialD when looking at some of the features in SQL Server 2012 - namely the Columnstore Index (for the new data warehousing crowd) and the new SSAS Tabular Model
>
>
>Both of those features use the xVelocity (formerly "Vertipaq") compression engine. That will likely compress integers better than any 32 character, 16 byte value. There are many vector-based efficiencies baked into the xVelocity engine that integers should leverage better than any type of uniqueidentifier. I'm going to see if I can write a sample in SQL 2012 to illustrate that. (I'm really curious know about what the difference would be)
>
>And thinking out loud, I wouldn't be surprised if the new in-memory optimized OLTP engine (Hekaton) in 2014 will work more efficiently with integer keys.

Would love to see some science instead of all the dogma. Here's a question, have you ever seen anyone use the entire 4 byte int range? Seems everyone forgets to use the negative numbers. I mean if the difference of 8 or 12 bytes is significant, then why cut the int in half by only using the positive numbers. That also seems like some kind of aversion, rather than science.