x
• Status: Solved
• Priority: Medium
• Security: Public
• Views: 636

# SMALLDATETIME Question

I have the following WHERE:

WHERE      liabhist.logtime = '4/23/02'

liabhist.logtime is a SMALLDATETIME and the WHERE does not work.  How do code the where so it will pick up all records from 4/23/02?
0
jasonboetcher
• 5
• 5
• 3
• +2
1 Solution

Commented:

WHERE Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) = '4/23/02'

This causes every logtime to look like '4/23/02 12:00:00 AM'  Which is what the other side oh the = sign looks like after it is converted too . . ..

0

Commented:
WHERE      liabhist.logtime >= '4/23/02' and liabhist.logtime < '4/24/02'
0

Author Commented:
Which of these suggestions is more efficient?
0

Commented:
WHERE Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) = '4/23/02' seems that it would be quite hard to read and by looking at it it seems that it would be doing a lot of calculations.  I just tried the queries in query analyzer on our invoice table about 1.5 million records I ran both queries together my way the other way.

My way took .07% of the total time to execute the other query with the convert took 99.93% of the time.
0

Senior .Net ConsultantCommented:
testing for equality (WHERE Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) = '4/23/02') is faster then the range method (WHERE      liabhist.logtime >= '4/23/02' and liabhist.logtime < '4/24/02' )
0

Commented:
it;s about the same.  Mine will cause slightly more CPU but less reads.  If your really curious run each query in Query analyzer like this

set statistics io on
set statistics time on

. . . .
WHERE      liabhist.logtime >= '4/23/02' and liabhist.logtime < '4/24/02'

Then run the other one in a different window . . .

set statistics io on
set statistics time on

. . . .
WHERE      liabhist.logtime >= '4/23/02' and WHERE Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) = '4/23/02'

And compare the reads and over all exec time . . . this will tell you for sure . . .
0

Senior .Net ConsultantCommented:
No need to convert to char before converting to date:

Convert(Datetime, liabhist.logtime,101) = '4/23/02'
0

Commented:
It is going to depend on how you have your table setup.  I am hitting my table on a clustered index.  I am returning 1300 records.  I would say the more records you have to go through the worse the Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) query is going to be because it is going to calculate that on every field.
0

Commented:
Without the extra convert that evens the execution time out to the same but the query plan has become more complicated.
0

Commented:
I agree.  The table set up is the key.  This all depends on what kind of indexes are on the table and their integrity/cardinality.  But in general the less reads the better.  It's a good idea to sacrifice CPU for reads.  CPU happens alot faster and doesn't strain your disk subsystem.
0

Senior DBACommented:
It's my understanding of SQL Server optimizer that it won't use an index for a comparison such as:

WHERE Convert(Datetime,Convert(VarChar,liabhist.logtime,101)) = '4/23/02'

but it will use one for:

WHERE liabhist.logtime >= ... AND < ...

On a large table, that could make a HUGE difference.  I don't think the CONVERT is that much overhead (it's just a little CPU anyway), but not using an index could be a killer.

By the way, if you do decide to run comparisons, be sure to do a:
DBCC DROPCLEANBUFFERS
to empty all buffers between runs.  Otherwise the second run should always be faster (the data needed has already been read into the buffer).
0

Commented:
I can run those queries any way and it will return the same results both ways.  I take the extra covert out and they are even I put it back in and is quite a bit slower. There is a definate balance between CPU and reads but the query took about 6 seconds for me to execute.  If I have a choice between 6 second and .5 second response.  You have to go with the .5 second.
0

Commented:
Well, I took my own advice and tested the IO and Exec times.  My query sucked wind compared to Curtis591's. there were about twice the reads.  I think ScottPletcher is right.  It doesn't seem to be using the index.  I even added in a with index table hint, but no change.  Interesting topic . . . =)
0

Senior DBACommented:
You might want to do a SHOWPLAN and see if the CONVERT version is doing some type of table/index scan versus the other method doing some type of index seek.  It would be interesting to know.
0

Commented:
It's doing an idex scan rather than a seek.  There was also a bookmark lookup that can be eliminated if I only had the indexed field in the select list (rather than Select *).
0

Senior DBACommented:
Thanks for checking on that.  It's interesting.  It's also about what I expected.

And obviously the larger the table, the more significant a SCAN vs. SEEK will be.  On a large table it's probably worth some extra code to be able to do the WHERE on the original column rather than a formula/derivation based on that column; do the manipulation on the local variables to be compared against the column, not the original column itself.
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.