如何优化T-SQL查询？

Question

I am trying to optimize the following query to increase the performance particularly the joins. I would greatly appreciate any suggestions and help. Thanks a lot. I am thinking to use CTE instead of subqueries

Query:

Select  year(dtb.tbDATE) as YearR, 
     `CONVERT(VARCHAR(12),dtb.tbDATE,110) as DateR,
     (case when dtb.l_grp_no= 7 then COMPANY_B else COMPANY_A end ) as Portfolio,
     (case when dtb.past_days between 5 and 30 then '5-30'
          when dtb.past_days between 31 and 60  then'31-60'
          when dtb.past_days between 61 and 90  then '61-90'
          when dtb.past_days >= 91 then '91+' 
          when dtb.past_days <5 then 'Current' else 'Dis_po' end) 'Qdel',
     case when lsc.SStatusC is not null and dtb.tbDATE> ls.eff_date 
          then lsc.SStatusC else 'XX'end as 'LStatus',
     count(dtb.refaccno) as lcount,
     sum (dtb.P_bal) as Lbal
from sln.[dbo].[table_dtb] as dtb
     --left outer join sln.dbo.acct_l la on dtb.accrefno = la.accrefno
     left outer join (select stat_acct_l.* from sln.dbo.stat_acct_l
     inner join 
         (select refaccno, max(row_id) as MaxRow_id
          from sln.dbo.stat_acct_l
          group by refaccno) as maxStatus 
        on stat_acct_l.refaccno = maxStatus.refaccno 
            and stat_acct_l.row_id = maxStatus.MaxRow_id) as ls 
        on ls.refaccno = dtb.refaccno
     left outer join dw.dbo.AccSt_C lsc on lsc.Stat_C_ID= ls.status_code_no
where dtb.l_grp_no in (7,4,8,15)and dtb.tbDATE > '2010-06-31'
     and dtb.P_bal+dtb.l_C_bal >0
group by year(dtb.tbDATE), CONVERT(VARCHAR(12),dtb.tbDATE,110),
     case when dtb.past_days between 5 and 30 then '5-30'
          when dtb.past_days between 31 and 60  then'31-60'
          when dtb.past_days between 61 and 90  then '61-90'
          when dtb.past_days >= 91 then '91+' 
          when dtb.past_days <5 then 'Current' else 'Dis_po' end
     case when dtb.l_grp_no= 7 then COMPANY_B else COMPANY_A end,
          case when lsc.SStatusC is not null 
                    and dtb.tbDATE> ls.eff_date Yearthen 
           lsc.SStatusC else 'XX'
      end
order by year(dtb.tbDATE), CONVERT(VARCHAR(12),dtb.tbDATE,110),
      case when dtb.l_grp_no= 7 then COMPANY_Belse COMPANY_A end

Answer 1

I noticed a couple of things. First, you can optimize your case constructs if you first take a look at your data. Then order your cases so that the ones that are most likely to return true are looked at first. You may have already done that, it's hard to tell.

Next, move more filtering from the where clause to the join clause. For example, this:

left outer join dw.dbo.AccSt_C lsc on lsc.Stat_C_ID= ls.status_code_no
WHERE dtb.l_grp_no in (7,4,8,15)and dtb.tbDATE > '2010-06-31' 
and dtb.P_bal+dtb.l_C_bal >0

might run faster like this:

left outer join dw.dbo.AccSt_C lsc on lsc.Stat_C_ID= ls.status_code_no
AND dtb.l_grp_no in (7,4,8,15)and dtb.tbDATE > '2010-06-31' 
and dtb.P_bal+dtb.l_C_bal >0

Also, don't select field you don't need. Do you really need every column here?

left outer join (select stat_acct_l.* from sln.dbo.stat_acct_l

Maybe these will help.

Answer 2

The more columns you can remove from the GROUP BY clause, the better. Depending on the number of rows you are dealing with, it may be worth it to make a temp table or table variable based on your select list that includes a column called something like "past_days_group" to hold 5-30, 31-60 etc.. Do the CASE logic that is currently in your group by clause when inserting to the temp table, then you can select your query out of the temp table grouping by "past_days_group".

Answer 3

What I would try is to encapsulate your case statements into UDFs. These should be calculated quicker as UDF than like regular case statements. This would also make your query easier to analyze and maintain.