Comments on: T-SQL Tuesday – Query Cost

By: Rob Farley

Rob Farley — Tue, 09 Mar 2010 15:12:05 +0000

Yes, I demonstrated that in my sqlbits.com talk, where a Compute Scalar had lots of hidden reads. stats io showed just 2 reads, when there were many in Profiler.
But the problem with this query is the intense CPU needed. There are some hidden I/O costs, but that’s a small impact compared to the CPU.

By: Mladen Prajdic

Mladen Prajdic — Tue, 09 Mar 2010 14:16:48 +0000

I’d run profiler to get the correct reads for your second method.
i’ve seen STATS IO get totally screwed up when doing string manipulations.
i have a query where the STATS IO show 30k reads while profiler shows 500k reads.

By: AaronBertrand

AaronBertrand — Tue, 09 Mar 2010 06:59:18 +0000

Yup, you’re really right. I abandoned my original idea though, and published a more broad "bad habits" type of I/O post.

By: Armando Prato

Armando Prato — Tue, 09 Mar 2010 06:52:36 +0000

I’d give SET STATISTICS TIME ON a shout out too. I’ve found my own fair share of instances where a query that appears to run acceptably time-wise and I/O wise may be over utilizing CPU.

By: Rob Farley

Rob Farley — Tue, 09 Mar 2010 02:51:50 +0000

You should still write it though.

By: AaronBertrand

AaronBertrand — Tue, 09 Mar 2010 02:38:52 +0000

Bummer! I was going to do a "Bad habits to kick : ignoring I/O" post for tomorrow, but my ideas were pretty much identical to yours. Kudos.