SQL Server: Order By DateDiff Performance issue - sql-server

Question

I'm having a problem getting top 100 rows from a table with 2M rows in reasonable time.
The problem is the order by part, it takes more than 50 minutes to get results for this query..
What can be the best solution for this problem?
select top 100 * from THETABLE TT
Inner join SecondTable ST on TT.TypeID = ST.TypeID
ORDER BY DATEDIFF(Day, TT.LastCheckDate, GETDATE()) * ST.SomeParam DESC
Many thanks,
Bentzy
Edit:
* TheTable is the one with 2M rows.
* SomeParam has 15 distinct values (more or less)

Answer 1

There are two things that come to mind to speed up this fetch:
If you need to run this query often, you should index the column 'lastCheckDate'. No matter which sql db you are using, a well defined index on the column will allow for faster selects, especially in an orders by clause.
Perform the date math before doing the select query. You are getting the difference in days between the row's checkDate and the current date, times some parameter. Does the multiplication affect the ordering of the rows? Can this simply be ordered by the 'lastCheckDate desc'? Explore other sorting options that return the same result.

Answer 2

Two ideas come to mind:
a) If ST.param doesn't change often, perhaps you can cache the result of the multiplication somewhere. The numbers would be "off" after a day, but the relative values would be the same - i.e., the sort order wouldn't change.
b) Find a way to reduce the size of the input tables. There are probably some values of LastCheckDate &/or SomeParam that will never be in the top 100. For example,
Select *
into #tmp
from THETABLE
where LastCheckDate between '2012-06-01' and getdate()
select top 100 *
from #tmp join SecondTable ST on #tmp.TypeID = ST.TypeID
order by DateDiff(day, LastCheckDate, getdate()) * ST.SomeParam desc
It's a lot faster to search a small table than a big one.

Answer 3

DATEDIFF(Day, TT.LastCheckDate, GETDATE()) is the number of days since "last check".
If you just order by TT.LastCheckDate you get a similar order.
EDIT
Maybe you can work out what dates you don't expect to get back and filter on them. Of course you then also need an index on that LastDateCheck column. If everything works out, you can at least shorten the list of records to check from 2M to some managable amount.

Answer 4

It is quite complicated.Do you seriouslly need all columns in query?There is one thing which you could try here. First just get the top 100 rows typeid
something like below
select top 100 typeid
,TT.lastcheckdate,st.someparam --do not use these if the typeid is unqiue in both tables..
--or just the PK columns of both tables and typeid so that these can be joined on PK
into #temptable
from st inner join tt on st.typeid = tt.typeid
ORDER BY DATEDIFF(Day, TT.LastCheckDate, GETDATE()) * ST.SomeParam DESC
Above will sort very minimal data and thus should be faster.Based on how many columns you have in your table and indexes this should be way faster (it will be fast if you have many columns in both tables but this query will use just 3.Also, maybe these columns (st.typeid,st.someparam and tt.typeid and tt.lastcheckdate) are covered by some of indexes so no need to read underlying tables and thus reduce the IO as well) than actual one..Then join this data back to both tables.
If that doesnt work the way you expect.Then you can have indexed view using above select by adding the order by expression as column. Then use this indexed view to get top 100 and join with main tables.This will surely reduce the amount of work and thus improve perf.But Indexed view will have overhead which will depend on how frequently data changed in the table TT.

Answer 5

To lessen number of rows you might retrieve top (100) for each SecondTable record ordered by LastCheckDate, and then union all them and finally select top (100), by means of temporary table or dynamic sql generated query.
This solution uses cursor to fetch top 100 records for each value in SecondTable. With index on (TypeID, LastCheckDate) on TheTable it runs instantaneously (tested on my system with a table of 700,000 records and 50 SecondTable entries).
declare #SomeParam varchar(3)
declare #TypeID int
declare #tbl table (TheTableID int, LastCheckDate datetime, SomeParam float)
declare rstX cursor local fast_forward for
select TypeID, SomeParam
from SecondTable
open rstX
while 1 = 1
begin
fetch next from rstX into #TypeID, #SomeParam
if ##fetch_status <> 0
break
insert into #tbl
select top 100 ID, LastCheckDate, #SomeParam
from TheTable
where TypeID = #TypeID
order by LastCheckDate
end
close rstX
deallocate rstX
select top 100 *
from #tbl
order by DATEDIFF(Day, LastCheckDate, GETDATE()) * SomeParam
Obviously this solution fetches ID's only. You might want to expand temporary table with additional columns.