I recently read an article on Bitbucket about “Don’t start identity columns or sequences with large negative values” and it got me wondering if the author’s conclusion is correct.
![](https://www.sqltopia.com/wp-content/uploads/2020/04/image.png)
So I set up a lab to test his hypothesis. Each table has about 4 million rows.
![](https://www.sqltopia.com/wp-content/uploads/2020/04/image-1.png)
And to mimic the property of IDENTITY, I then rebuilt the indexes on the tables.
![](https://www.sqltopia.com/wp-content/uploads/2020/04/image-2.png)
And then I executed these statements to see if there was a difference in compression, as the author stated.
![](https://www.sqltopia.com/wp-content/uploads/2020/04/image-3.png)
And as I suspected, there was no difference.
The reason is how page- and row-compression is implemented.
See whitepapers here
And to be fair, I did the test again with ROW compression instead of PAGE compression and here is the result
![](https://www.sqltopia.com/wp-content/uploads/2020/04/image-4.png)
Again, there is no difference. The miniscule 8KB is within the error margin.