Comments on: Seek or scan?

By: Performace Tuning Stored Procedures « SQL Canada

Performace Tuning Stored Procedures « SQL Canada — Wed, 21 Dec 2011 15:50:09 +0000

[...] Link. [3] Bell Curve. Wikipedia. Link. [4] Seek or Scan?. Microsoft SQL Server MVP, Gail Shaw. Link. [5] Dissecting SQL Server Execution Plans, Grant Fritchey. Link. Rate this: Share [...]

By: NInja Rg'Rus

NInja Rg'Rus — Mon, 15 Aug 2011 16:52:50 +0000

A very interesting post on the same subject by Kimberly Tripp.

http://www.sqlskills.com/BLOGS/KIMBERLY/post/The-Tipping-Point-Query-Answers.aspx

By: Jason

Jason — Wed, 16 Jun 2010 21:49:35 +0000

Any thoughts on this related post on Stack Overflow? Mike M. referenced this article, but at this time of writing, we’re still a little confused …

http://stackoverflow.com/questions/3056321/sql-2005-indexed-queries-slower-than-unindexed-queries

By: Gail

Gail — Thu, 16 Apr 2009 08:10:57 +0000

No. It’s because the ID column is part of that nonclustered index. ID is the clustered index key and hence is part of all nonclustered indexes because it’s the row’s address. (see – http://sqlinthewild.co.za/index.php/2009/02/09/all-indexes-are-unique/)

So your query with ID and SomeNumber was reading a covering nonclustered index.

By: Cedric

Cedric — Wed, 15 Apr 2009 23:56:23 +0000

Sorry I should have specifically mentioned that I was refering to the index defined as : CREATE NonCLUSTERED INDEX [idx_SeekOrScan] ON [dbo].[SeekOrScan] (SomeNumber) my tiping point seems to different than yours.I need more information in this regard.I will do some more investigation .

Agreed that if you add the INCLUDE (id, padding) that a non clustered index seek will be used.

I should have maybe worded my post differently . If I include the SomeNumber field explicitly in my select list as the only field then the non clustered index seems to be covering even if I add the ID field it still used the non clustered index seek ,but as soon as I add the padding field the query plan swings around to a clustered index scan

Does all this mean that depeinding on your select list that the tipping point can potentially be different if you explicitly state your field list ?

By: Gail

Gail — Wed, 18 Mar 2009 14:41:04 +0000

Cedric, are you saying that if I create an index (SomeNumber) INCLUDE (id, padding) and then run “select * from [SeekOrScan] where somenumber between 1 and 50000″ that I’ll still get a clustered index scan?

Sure, if you just specify the indexes column in the select it will be an index seek, because no bookmark lookup is required, the index alone can satisfy the query. It’s only when there’s a bookmark lookup that there’s a ‘tipping point’

By: Cedric

Cedric — Wed, 18 Mar 2009 12:14:29 +0000

Even creating a covering index in this instance will still yield a clustered index scan if the columns on which the covering index are defined is not part explicitly stated in the SELECT clause ,until the point where the ratio of rows like Gail illustrated above is reached.

If we changed one of the statements above to :

select somenumber from [SeekOrScan]
where somenumber between 1 and 50000

Then we see a non clustered index seek

By: Now syndicating Gail Shaw and Kendal Van Dyke | Brent Ozar - SQL Server DBA

Now syndicating Gail Shaw and Kendal Van Dyke | Brent Ozar - SQL Server DBA — Tue, 10 Feb 2009 23:11:12 +0000

[...] Seek or scan? – how does SQL Server determine when to pick one over the other? [...]

By: Christopher Stobbs

Christopher Stobbs — Mon, 26 Jan 2009 09:10:15 +0000

he he he Yip SP1, I’ve tried it on another machine and it all seems good thanks for that

By: Gail

Gail — Fri, 23 Jan 2009 16:26:35 +0000

Let me guess, SQL 2005 RTM or SP1?

The operator was called a bookmark lookup in SQL 2000 and before. In SQL 2005 SP2 and later it was a key lookup.

In SQL 2005 RTM and SP1 it appeared as a clustered index seek, and hence often went unnoticed (because clustered index seeks are good)