How many queries, percentage wise, would actually be ordered by an
identity column ... such that the noted technique would be valid?
Actually, you can make this technique value in all cases, and it's actually the only way to get repeatable results back.
Say you're using the select from select from select from... approach over a customer table, and you're sorted on lastname, firstname. It
is possible, even likely, that you've got a number of Mike Smiths in
the database. Your order by would be "order by lastname, firstname"
yes? But what order are the duplicate Mike Smiths going to show up
in? Since you didn't specify then the order is not defined, and there
is _NO_ guarantee that they will come back in the same order next
time, so the Mike Smith with customer # 12345 may have been the
1001st row first time and missed the 1-1000 page and been the 999th
row the second time and missed the 1001-2000 page. Therefore...
Make your order by "order by lastname, firstname, customer#" and then
have the <ed> caller pass in the last customer # they received
(@LastNumber). Then you do this sort of thing:
select @lastname = LastName, @FirstName=FirstName from customer where
select top 1000 * from customer where (lastname=@LastName and FirstName=@FirstName and
or (lastName=@LastName and FirstName>@FirstName)