SQL MAX() function is very slow, is there a better method? - asp.net

Question

I have a SQL database and ASP.NET web application and most of my queries involve the SQL max function.
For example, the below query takes between approx. 36 seconds to execute (when using the profiler) on both the ASP.NET website and the SSMS.
SELECT MAX(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) -
MIN(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) AS ACTUALHOURSRUN
FROM REPORTINGSYSTEM.DBO.HL_LOGS
WHERE ID_LOCATION = #ID_LOCATION AND
CONVERT(VARCHAR,TIME_STAMP,102)
BETWEEN #STARTDATE AND #ENDDATE
The table in question has approx. 5,000,000 records and 45 columns.
What is the best/fastest/most efficient way of executing the query to reduce the execution time?
Thanks in advance...

Answer 1

An index on ID_LOCATION and TIME_STAMP would be a good candidate.
However, as you are converting the TIME_STAMP field to a VARCHAR, this will prevent it from being able to use the index. Instead, rework the query to remove the need to do the CONVERT and instead work with DATETIME values.
e.g.
SELECT MAX(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) - MIN(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) AS ACTUALHOURSRUN
FROM REPORTINGSYSTEM.DBO.HL_LOGS
WHERE ID_LOCATION = #ID_LOCATION AND TIME_STAMP >= #STARTDATE AND TIME_STAMP < #ENDDATE
Ensure #STARTDATE and #ENDDATE are DATETIME parameters also (I'm assuming this is the datatype of the TIME_STAMP column)
I'd also question what datatypes are the Runhrs/Runho columns? If not stored as FLOAT/DECIMAL already, then could they not be as that would appear to be the most suitable datatype?

Answer 2

There are several things that you need to do:
Make sure that the columns that you search are indexed - You need an index on ID_LOCATION and TIME_STAMP (if you have other queries that query by date across locations or by locations across dates, you may consider defining separate indexes; otherwise, a single combined index will work).
Stop converting the timestamp to a string - use the native data type for #STARTDATE and #ENDDATE in your query, or replace the condition with TIME_STAMP between CONVERT(datetime,#STARTDATE) and CONVERT(datetime,#ENDDATE)
These two changes should make your query faster, especially the second one: currently, CONVERT(VARCHAR,TIME_STAMP,102) forces the query optimizer into a full scan on everything that matches your location, or even a full table scan if there is no index on ID_LOCATION. An indexed search should bring the number of records down to acceptable levels.
The final item to take on is CONVERT(FLOAT,ISNULL(Runhrs, Runho)): if the query speed after the first two modifications remains insufficient, change MAX(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) to CONVERT(FLOAT, MAX(ISNULL(Runhrs,Runho))), and do the same change for the MIN. This may or may not work, depending on the types of Runhrs and Runho.

Answer 3

First, store Rhnhrs as a numeric type. Then you don't need to do the conversion.
Second, you'll make this faster by creating an index on hl_logs(id_location, time_stamp). You could also throw Runhrs and Runho as well. With an index on all four columns (in that order), the query will not even need to go to the original data.
To use the index, you need to change the where statement to something like:
time_stamp bewteen #startTimeStamp and #EndTimeStamp
The SQL engine won't use the index if the variable is the argument to a function.
The resulting query should look more like:
select max(coalesce(runhrs, runho)) - min(coalesce(runhrs, runho) as Actual
from REPORTINGSYSTEM.DBO.HL_LOGS
WHERE ID_LOCATION = #ID_LOCATION AND
TIME_STAMP BETWEEN cast(#STARTDATE as datetime) AND cast(#ENDDATE as datetime)

Answer 4

SELECT CONVERT(FLOAT, MAX(Runhrs)), CONVERT(FLOAT, MAX(Runho),
CONVERT(FLOAT, MIN(Runhrs)), CONVERT(FLOAT, MIN(Runho)
FROM REPORTINGSYSTEM.DBO.HL_LOGS
WHERE ID_LOCATION = #ID_LOCATION
AND TIME_STAMP BETWEEN #STARTDATE AND #ENDDATE
Do the subtraction yourself in the code which call the SQL. (That is the .net code.)
You should have an index on ID_LOCATION, TIME_STAMP, Runhrs, Runho. One index with all four fields. Maybe two indexes.
CREATE INDEX REPORTINGSYSTEM.DBO.HL_LOGS ON ID_LOCATION, TIME_STAMP, Runhrs
CREATE INDEX REPORTINGSYSTEM.DBO.HL_LOGS ON ID_LOCATION, TIME_STAMP, Runho

Answer 5

AND CONVERT(VARCHAR,TIME_STAMP,102)
This is NOT the way to do it - it will cause slowness. Do NOT apply unnecessary functions to the DATA - it robs the optimizer of any chance to use indexes on this field. Do it the other way around, convert the #STARTDATE + #ENDDATE into the same data type as the field. (or, think of it this way you are doing ~ 5,000,000 conversions so that you can compare to 2 variables)
[time_stamp] BETWEEN #STARTDATE AND #ENDDATE
Please understand this actually means:
[time_stamp] > = #STARTDATE AND [time_stamp] <= #ENDDATE
i.e. BOTH "boundary" datetimes are INCLUDED, usually it's easier and more accurate to avoid between's inclusive nature and spell it out like this:
[time_stamp] > = #STARTDATE AND [time_stamp] < #ENDDATE (with #enddate
being "next day at 00:00:00:000)
MAX(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) -
MIN(CONVERT(FLOAT,ISNULL(Runhrs,Runho))) AS ACTUALHOURSRUN
wow, This is a lot of function calls! It's not pure MAX()/MIN() that's at issue here.
Suggest you simply look for Max(Runhrs) etc as previously mentioned, THEN do the subtractions.