Question

I have one general question about optimization at SQL Server: How I can reduce number of Reads in stored procedure?

I’m interested in good practices like:
– creating indexes at physical and temp tables
– using temp tables instead using same table couple times in procedure
– DDL before DML
– SET NOCOUNT ON at the beginning of stored procedure
– …

We have issue with disk space that is used because of huge number of reads that is caused by couple stored procedures and I need to optimize it.

Part of stored procedure that is ‘the most expensive’ is:

create table #stavke   
(  
    Id_Br int identity(1, 1), IDStavke int, 
    HeaderID int, Currency varchar(3),  GLAcct varchar(20), id varchar(20), Trnuid varchar(60), 
    ReferenceID varchar(20), DocumentID varchar(20),
    DtAvail varchar(10), DtBooking varchar(10), DatePosted varchar(10),
    Amount money, AmountLcl money, 
    Description varchar(250), Type varchar(10), DP int  )
insert into #stavke   
  (  
    IDStavke, HeaderID, GLAcct, Currency, id , Trnuid , 
    ReferenceID, DocumentID ,
    DtAvail , DtBooking , DatePosted,
    Amount , AmountLcl , 
    Description , Type , DP 
  )  

SELECT S.ID as IDStavke,
   z.RB as HeaderID,  
   z.KONTO AS GLAcct,
   z.OZNVAL AS Currency,
   Si.BROJNALOGA as ID,
   D.TRN as Trnuid, 
   case substring(SI.BROJNALOGA,1,4)
        when '0746' then O.REFERENCA
        when '1450' then D.REFERENCA
        when '0743' then L.REFERENCA
        when '2021' then N.REFERENCA
   end   ReferenceID,
   case substring(SI.BROJNALOGA,1,4)
        when '3000' then 'Kursna razlika'
        when '2200' then 'PP-'+SI.BROJNALOGA
        when '2201' then 'KDP-'+SI.BROJNALOGA
        else SI.BROJNALOGA
   end DocumentID,
   dvalute as DtAvail, 
   si.dknizenja as DtBooking, 
   '' as DatePosted,        
   case si.teret
        when 0 then si.korist
        else si.teret 
   end Amount, 
   case SI.DINTERET
        when 0 then si.dinkorist
        else si.dinteret 
   end AmountLcl, 
   '' as Description,
   case substring(SI.BROJNALOGA,1,4)
        when '0746' then '0746'
        when '1450' then '1450'
        when '0743' then '0743'
        when '2021' then  'Ostalo'
   end  Type,
   case SI.DINTERET
        when 0 then 1
        else -1 
   end DP
FROM       A I
inner join B st on i.transfer=st.transfer and i.partija=st.partija 
INNER JOIN C SI ON st.RB=Si.RB
inner join D z on z.rb=st.rb
inner join E s on z.rb=s.rb AND s.BROJNALOGA = si.BROJNALOGA 
LEFT JOIN  F D ON  D.BROJ=SI.BROJNALOGA
LEFT JOIN  G L ON L.BROJ=SI.BROJNALOGA
LEFT JOIN  H O ON O.BROJ=SI.BROJNALOGA
LEFT JOIN  I N ON N.BROJ=SI.BROJNALOGA 
WHERE I.novi_izvod=convert(int,@StatementNumber) AND i.PARTIJA=@Account 
ORDER BY I.PARTIJA,z.RB,SI.id, z.KONTO,z.OZNVAL, SI.DKNIZENJA

Tables B, G, H and I (I changed real names of tables for this example to make it easier for reading) are very big, i.e. have many columns and big amount of data.

Answer 1

I hope you are talking of minimizing the disk activity by a procedure.

firstly, you can benchmark your current IO activity using the

set statistics IO on;

With that info at hand and getting the Execution Plan with SET SHOWPLAN_ALL or XML or you can make use of ssms to get the same ergonomically. You can make use of DTA for basic tuning.

Try executing the SP as a bunch of ad-hoc statements and see where the IO is heavy and concentrate on that segment. There are a lot of good practices which might suit your requirement.