Click here to monitor SSC
SQLServerCentral is supported by Red Gate Software Ltd.
 
Log in  ::  Register  ::  Not logged in
 
 
 
        
Home       Members    Calendar    Who's On


Add to briefcase

UDF Performance When Used As A "Macro" Expand / Collapse
Author
Message
Posted Friday, October 12, 2012 7:22 AM
SSC Rookie

SSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC Rookie

Group: General Forum Members
Last Login: Monday, June 9, 2014 7:00 AM
Points: 37, Visits: 112
My group is having a discussion as to performance hits when a UDF is used as a macro -
IE it accesses no tables, is passed no rows. Some simple logic and a calculation on a date/time
value.

Here is an abbreviated version of the code (removed check for DST):


Create FUNCTION fnAdjustDateTime (@date datetime, @timezone int)

RETURNS datetime
AS
BEGIN
--
-- declare variables
--
declare @dst bit
declare @offset int
declare @result datetime
--
-- Get the timezone data we will need
--
if @timezone = 1
Select @offset = -5, @dst = 1
else if @timezone = 2
Select @offset = -6, @dst = 1
else if @timezone = 3
Select @offset = -7, @dst = 1
else if @timezone = 4
Select @offset = -8, @dst = 1
else if @timezone = 5
Select @offset = -10, @dst = 0
else if @timezone = 6
Select @offset = -9, @dst = 1
else if @timezone = 7
Select @offset = -4, @dst = 0
else if @timezone = 8
Select @offset = -7, @dst = 0
else if @timezone = 9
Select @offset = -11, @dst = 0
else
Select @offset = -6, @dst = 1
--
-- Set result now
--
set @result = DateAdd(hh,@offset,@date)

return @result




Could someone please provide some insight on this as well a a logical explanation ??

Thanks ...

Todd


Post #1372111
Posted Friday, October 12, 2012 7:37 AM


SSChampion

SSChampionSSChampionSSChampionSSChampionSSChampionSSChampionSSChampionSSChampionSSChampionSSChampion

Group: General Forum Members
Last Login: Yesterday @ 5:47 PM
Points: 12,910, Visits: 32,020
well, ideally, your scalar UDF can be converted to an inline table funciton instead.
Create  FUNCTION fnAdjustDateTime_itv (@date datetime, @timezone int)

RETURNS table
AS
RETURN
SELECT DateAdd(hh,CASE
WHEN @timezone = 1 THEN -5
WHEN @timezone = 2 THEN -6
WHEN @timezone = 3 THEN -7
WHEN @timezone = 4 THEN -8
WHEN @timezone = 5 THEN -10
WHEN @timezone = 6 THEN -9
WHEN @timezone = 7 THEN -4
WHEN @timezone = 8 THEN -7
WHEN @timezone = 9 THEN -11
END,
@date) As ResultDate
GO
Create FUNCTION fnAdjustDateTime (@date datetime, @timezone int)

RETURNS datetime
AS
BEGIN
--
-- declare variables
--
declare @dst bit
declare @offset int
declare @result datetime
--
-- Get the timezone data we will need
--
if @timezone = 1
Select @offset = -5, @dst = 1
else if @timezone = 2
Select @offset = -6, @dst = 1
else if @timezone = 3
Select @offset = -7, @dst = 1
else if @timezone = 4
Select @offset = -8, @dst = 1
else if @timezone = 5
Select @offset = -10, @dst = 0
else if @timezone = 6
Select @offset = -9, @dst = 1
else if @timezone = 7
Select @offset = -4, @dst = 0
else if @timezone = 8
Select @offset = -7, @dst = 0
else if @timezone = 9
Select @offset = -11, @dst = 0
else
Select @offset = -6, @dst = 1
--
-- Set result now
--
set @result = DateAdd(hh,@offset,@date)

return @result
END
GO

then the calling code gets changed to this:

select name, dbo.fnAdjustDateTime(create_date,1) from sys.objects

select name, fn.*
from sys.objects
CROSS APPLY dbo.fnAdjustDateTime_itv(create_date,1) fn


if you include actual execution palns and compare them, you will start seeing hwo the ITVF version is better; scalar functions usualyl get called once per row,
where an ITVF is treated like a set based operation

on this small dataset of using sys.objects, my results were the same; but on bigger datasets, the ITVF will be much more efficient.


Lowell

--There is no spoon, and there's no default ORDER BY in sql server either.
Actually, Common Sense is so rare, it should be considered a Superpower. --my son
Post #1372124
Posted Friday, October 12, 2012 8:25 AM


SSC-Dedicated

SSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-Dedicated

Group: General Forum Members
Last Login: 2 days ago @ 9:58 AM
Points: 36,995, Visits: 31,517
Todd Young-359443 (10/12/2012)
My group is having a discussion as to performance hits when a UDF is used as a macro -
IE it accesses no tables, is passed no rows. Some simple logic and a calculation on a date/time
value.


To add to what Lowell has posted, "One test is worth a thousand expert opinions". Please see the following article for what a difference an "iSF" can make over such a Scalar UDF.
http://www.sqlservercentral.com/articles/T-SQL/91724/


--Jeff Moden
"RBAR is pronounced "ree-bar" and is a "Modenism" for "Row-By-Agonizing-Row".

First step towards the paradigm shift of writing Set Based code:
Stop thinking about what you want to do to a row... think, instead, of what you want to do to a column."

(play on words) "Just because you CAN do something in T-SQL, doesn't mean you SHOULDN'T." --22 Aug 2013

Helpful Links:
How to post code problems
How to post performance problems
Post #1372171
Posted Monday, October 15, 2012 11:46 AM
SSC Rookie

SSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC RookieSSC Rookie

Group: General Forum Members
Last Login: Monday, June 9, 2014 7:00 AM
Points: 37, Visits: 112

I was able to work at home this weekend and found the answer. When calling
at UDF that does not access any database table and is used more like a macro,
the performance hit is negligible.

The raw difference in a 20,000 row dataset was 200 milliseconds. When I removed
the overhead of the three DateAdd functions it dropped to 100 milliseconds.

We are keeping the function in production and have dropped this as a candidate
for optimization.



Post #1372857
Posted Monday, October 15, 2012 12:09 PM


SSC-Dedicated

SSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-DedicatedSSC-Dedicated

Group: General Forum Members
Last Login: 2 days ago @ 9:58 AM
Points: 36,995, Visits: 31,517
Todd Young-359443 (10/15/2012)

I was able to work at home this weekend and found the answer. When calling
at UDF that does not access any database table and is used more like a macro,
the performance hit is negligible.

The raw difference in a 20,000 row dataset was 200 milliseconds. When I removed
the overhead of the three DateAdd functions it dropped to 100 milliseconds.

We are keeping the function in production and have dropped this as a candidate
for optimization.

I guess it really depends on what you call "negligible". You're talking about a function that runs twice as fast and that's not "negligible" especially when you consider that you're only working with 20,000 rows.

"Mind the pennies and the dollars will take care of themselves."


--Jeff Moden
"RBAR is pronounced "ree-bar" and is a "Modenism" for "Row-By-Agonizing-Row".

First step towards the paradigm shift of writing Set Based code:
Stop thinking about what you want to do to a row... think, instead, of what you want to do to a column."

(play on words) "Just because you CAN do something in T-SQL, doesn't mean you SHOULDN'T." --22 Aug 2013

Helpful Links:
How to post code problems
How to post performance problems
Post #1372868
Posted Tuesday, October 16, 2012 7:24 AM


SSCarpal Tunnel

SSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal TunnelSSCarpal Tunnel

Group: General Forum Members
Last Login: Yesterday @ 7:50 PM
Points: 4,351, Visits: 6,167
Todd Young-359443 (10/15/2012)

I was able to work at home this weekend and found the answer. When calling
at UDF that does not access any database table and is used more like a macro,
the performance hit is negligible.

The raw difference in a 20,000 row dataset was 200 milliseconds. When I removed
the overhead of the three DateAdd functions it dropped to 100 milliseconds.

We are keeping the function in production and have dropped this as a candidate
for optimization.



You are fooling yourself. There are MANY reasons why UDFs are bad (see my chapter in the SQL Server MVP Deep Dives 2 book titled Death By UDF and my SQL Rally 2012 session of the same name). Depending on how this UDF gets used it could prevent the optimizer from getting accurate statistics on the query and thus lead to a disastrously bad query plan. And did you know that UDFs also void the use of parallelism? Lots of other potential issues come with them.


Best,

Kevin G. Boles
SQL Server Consultant
SQL MVP 2007-2012
TheSQLGuru at GMail
Post #1373243
« Prev Topic | Next Topic »

Add to briefcase

Permissions Expand / Collapse