I have a table in a database with hundreds of millions of rows. The records are quite small (30 bytes) with a clustered index on TagID (Int) and SampleDateTime (BigInt). I didn't design the table but I am almost convinced the the clustered index should be reversed because SampleDataTime has much greater uniqueness, always increases (obviously) and we ALWAYS specify a DateTime range in our queries.

I experimented and found that the execution time for queries that specify a DateTime range but no TagIDs came down from 45 seconds to less than 1 second. The only problem is that most existing queries take much longer, presumably because they are structured for the original clustered index,