Comments on: How to make text searching go faster

By: Rob Farley

Rob Farley — Sat, 26 Mar 2016 04:12:15 +0000

No worries, and thanks for supporting it.

By: Saeid Hasani

Saeid Hasani — Sat, 26 Mar 2016 04:03:04 +0000

Thanks Rob for your fast response. By referring to that article , I wanted to confirm your solution. I’m not against Uri’s idea. He is a bright person in SQL world. But, this solution has its opponents (like few comments on my article). So, I was become happy when I read your great post. I didn’t consider Brent’s article references. I downloaded your SQLBits session and I saw it . It is an awesome session. Wish I saw it few years ago! 🙂 Thank you so much!

By: Rob Farley

Rob Farley — Sat, 26 Mar 2016 03:16:53 +0000

Yes, Saeid. Lots of situations. Nice article - I wasn't aware of it. I presented this as one of my T-SQL Tips at TechEd AU back in 2008, but it wasn't new then either. Interesting that you refer to Brent's article on Sargability - he refers to one of mine in his, and was in the audience for my SQLBits session on it in 2010, which you can see at http://bit.ly/Sargability

By: Saeid Hasani

Saeid Hasani — Sat, 26 Mar 2016 02:54:10 +0000

There are many situation that we can use this method such as codes, barcodes, national security codes etc. http://social.technet.microsoft.com/wiki/contents/articles/31238.t-sql-improve-the-performance-for-like-wildcard-by-changing-the-collation.aspx

By: Rob Farley

Rob Farley — Wed, 09 Mar 2016 10:06:56 +0000

Yes. If you need to find "Uri" and "URI" and "uri", then you can’t use a binary collation. But if it’s always stored as "URI", then you can.

By: Uri Dimant

Uri Dimant — Wed, 09 Mar 2016 09:56:48 +0000

I am aware of this Rob.
Using binary collation as you pointed out may return incorrect data (not all data found)

By: Rob Farley

Rob Farley — Tue, 08 Mar 2016 08:19:15 +0000

You mean the question about why a Hash is 10% (ish) faster in a binary collation?
It’s because with a non-binary collation, it needs to make a call to GetSortKey(), so that it can hash a value which is equal for all the things that sort to the same point in that collation. Like how in a CI collation, H and h need to end up in the same hash bucket. In a binary collation, it doesn’t have to call GetSortKey(), because things are only equal when they have exactly the same binary value.
I worked this out by asking someone who had access to a debugger which they regularly hook SQL Server into. He’s a good friend, and happens to live only a few time zones away from me. (Thanks Paul!)

By: Uri Dimant

Uri Dimant — Tue, 08 Mar 2016 07:56:54 +0000

Hi Rob
It is great post, but question is not answered :-))) Waiting for your investigation.