IDS Forum

Re: Slow Query with index

Posted By: Laurie Gustin
Date: Tuesday, 10 May 2011, at 2:49 p.m.

In Response To: Re: Slow Query with index (Art Kagel)

Thanks for the insight Art!

Still not knowing exactly why this queries performs differently, I took what you said and put it all in a stored procedure that will separate the 'two sides of the union' and only do the second half if needed. It returns a sub-second response - no matter what part of the alphabet you are in.. or how many records for a single last name.

Thanks for everyone's help on this!!

Laurie

Laurie Gustin
Database Administrator
Dept of Technology Services
Dept of Public Safety
lgustin@utah.gov
801-965-4410

>>> "Art Kagel" <art.kagel@gmail.com> 5/10/2011 11:17 AM >>>
OK, here's the problem. The optimzer is treating the inequality filter and
the equality filters as if this were a UNION of two separate queries. So it
is seeing:

select first 100

vo_veh_id, vo_owner_type, vo_first_name, vo_last_name,
vo_middle_name,

vo_addr_set_id,vehownindex
from vehicle_owner
where ((vo_owner_type = 'O' OR vo_owner_type = 'E')

and ((vo_last_name = 'WADSWORTH' and vo_first_name >= 'RALPH')))
UNION
select first 100

vo_veh_id, vo_owner_type, vo_first_name, vo_last_name,
vo_middle_name,

vo_addr_set_id,vehownindex
from vehicle_owner
where ((vo_owner_type = 'O' OR vo_owner_type = 'E')

and ((vo_last_name > 'WADSWORTH')))
order by vo_last_name, vo_first_name;

In order to join the two results sets, remove duplicates, and filter for the
first 100 rows, the optimizer must collect all of the matching rows from
both queries and sort the results. I'm betting that 'ANDERSON' is either
the smallest value in vo_last_name or there are more than 100 rows with that
value in the column and the optimizer knows that, so it is shortcutting the
query. It may even be eliminating the second part of the UNION or merging
the two into a single '<=' comparison if you also eliminated the filter on
vo_first_name when you substituted 'ANDERSON'. I don't know because you did
not supply the query plan for that version of the query. If you eliminate
the FIRST 100 clause, does it run faster for 'ROGER' & 'WADSWORTH'?

Art

Art S. Kagel
Advanced DataTools (www.advancedatatools.com)
Blog: http://informix-myview.blogspot.com/

Disclaimer: Please keep in mind that my own opinions are my own opinions and
do not reflect on my employer, Advanced DataTools, the IIUG, nor any other
organization with which I am associated either explicitly, implicitly, or by
inference. Neither do those opinions reflect those of other individuals
affiliated with any entity with which I am affiliated nor those of the
entities themselves.

On Tue, May 10, 2011 at 12:32 PM, Laurie Gustin <lgustin@utah.gov> wrote:

> Thanks for the suggestions...
>
> I changed the index, leaving out middle_name altogether and putting type at
> the end. Then ran the query without the index directive... below is the
> query
> plan. Unfortunately, I no longer have access to the old server to compare
> the
> old query plan...
>
> QUERY: (OPTIMIZATION TIMESTAMP: 05-10-2011 10:22:31)
> ------
> select
> first 100
> vo_veh_id, vo_owner_type, vo_first_name, vo_last_name, vo_middle_name,
> vo_addr_set_id,vehownindex
> from vehicle_owner
> where ((vo_owner_type = 'O' OR vo_owner_type = 'E') and
> ((vo_last_name = 'WADSWORTH' and vo_first_name >= 'RALPH') or (vo_last_name
> >
> 'WADSWORTH')))
> order by vo_last_name, vo_first_name
>
> Estimated Cost: 751789
> Estimated # of Rows Returned: 481854
> Temporary Files Required For: Order By
>
> 1) informix.vehicle_owner: INDEX PATH
>
> (1) Index Name: informix.idx_vehowner
>
> Index Keys: vo_last_name vo_first_name vo_owner_type (Key-First) (Serial,
> fragments: ALL)
>
> Lower Index Filter: informix.vehicle_owner.vo_last_name > 'WADSWORTH'
>
> Index Key Filters: ((informix.vehicle_owner.vo_owner_type = 'O' OR
> informix.vehicle_owner.vo_owner_type = 'E' ) )
>
> (2) Index Name: informix.idx_vehowner
>
> Index Keys: vo_last_name vo_first_name vo_owner_type (Key-First) (Serial,
> fragments: ALL)
>
> Lower Index Filter: (informix.vehicle_owner.vo_last_name = 'WADSWORTH' AND
> informix.vehicle_owner.vo_first_name >= 'RALPH' )
>
> Index Key Filters: ((informix.vehicle_owner.vo_owner_type = 'O' OR
> informix.vehicle_owner.vo_owner_type = 'E' ) )
>
> The query still took over 2 minutes.
>
> When I run the query with the last name of ANDERSEN, it takes about 1
> second...
>
> Thanks
> Laurie
>
> >>> "Art Kagel" <art.kagel@gmail.com> 5/10/2011 9:13 AM >>>
> Try adding this index and removing the +INDEX() optimizer directive:
>
> create index 'informix'.idx_vehowner on 'informix'.vehicle_owner
> ( vo_last_name asc,
> vo_first_name asc,
> vo_owner_type,
> vo_middle_name asc
> ) ;
>
> If that doesn't work, post the original and new SET EXPLAIN outputs and
> we'll take a look.
>
> Art
>
> Art S. Kagel
> Advanced DataTools (www.advancedatatools.com)
> Blog: http://informix-myview.blogspot.com/
>
> Disclaimer: Please keep in mind that my own opinions are my own opinions
> and
> do not reflect on my employer, Advanced DataTools, the IIUG, nor any other
> organization with which I am associated either explicitly, implicitly, or
> by
> inference. Neither do those opinions reflect those of other individuals
> affiliated with any entity with which I am affiliated nor those of the
> entities themselves.
>
> On Tue, May 10, 2011 at 11:00 AM, Laurie Gustin <lgustin@utah.gov> wrote:
>
> > IDS 11.50 FC8X4
> > Linux
> >
> > I have a query that has been successfully running on an IDS 10.0 for
> quite
> > some time. We recently moved the database to an 11.5 server, and the
> result
> > set appears different. For some reason the IDS 10 database seemed to
> return
> > the data in index order, even tho the index was not clustered. The 11.5
> > database did not do that, so when we limited the number of rows returned,
> > often the rows we were looking for were not included in the result set.
> To
> > fix
> > this, I added the 'order by' statement. The result set is now as
> expected,
> > however, the query takes up to 2 minutes to complete. The index is as
> > indicated.. and update statistics have been run using Art's dostats.
> >
> > select {+INDEX (idx_vehowner, vehicle_owner)}
> > first 100
> > vo_veh_id, vo_owner_type, vo_first_name, vo_last_name, vo_middle_name,
> > vo_addr_set_id,vehownindex
> > from vehicle_owner
> > where ((vo_owner_type = 'O' OR vo_owner_type = 'E') and
> > ((vo_last_name = 'WILSON and vo_first_name >= 'RUDOLPH') or (vo_last_name
> >
> > 'WILSON')))
> > order by vo_last_name, vo_first_name;
> >
> > create index 'informix'.idx_vehowner on 'informix'.vehicle_owner
> > ( vo_owner_type,
> > vo_last_name asc,
> > vo_first_name asc,
> > vo_middle_name asc
> > ) ;
> >
> > The really strange thing is that if I use a name beginning with 'A', the
> > query
> > is screaming fast. The farther into the alphabet you go, the slower it
> gets
> > -
> > kind of like it is just scanning through the table in order?
> >
> > there are 7.8 million records in the table.
> >
> > Any suggestions would be helpful!
> >
> > Thanks!
> > Laurie
> >
> > Laurie Gustin
> > Database Administrator
> > Dept of Technology Services
> > Dept of Public Safety
> > lgustin@utah.gov
> > 801-965-4410
> >
> >
> >
> >
>
>
>
*******************************************************************************
> > Forum Note: Use "Reply" to post a response in the discussion forum.
> >
> >
>
> --20cf3071d02e3dad8804a2ed66ea
>
>
>
>
*******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>
>
>
*******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>

--bcaec54865d22a809b04a2ef21d1

*******************************************************************************
Forum Note: Use "Reply" to post a response in the discussion forum.

Messages In This Thread

• Slow Query with index
  Laurie Gustin -- Tuesday, 10 May 2011, at 11:00 a.m.
  • Re: Slow Query with index
    Keith Simmons -- Tuesday, 10 May 2011, at 11:11 a.m.
  • Re: Slow Query with index
    Art Kagel -- Tuesday, 10 May 2011, at 11:13 a.m.
    • Re: Slow Query with index
      Laurie Gustin -- Tuesday, 10 May 2011, at 12:32 p.m.
      • Re: Slow Query with index
        Jack Parker -- Tuesday, 10 May 2011, at 12:56 p.m.
        • Re: Slow Query with index
          Laurie Gustin -- Tuesday, 10 May 2011, at 1:14 p.m.
      • Re: Slow Query with index
        Jack Parker -- Tuesday, 10 May 2011, at 1:17 p.m.
      • Re: Slow Query with index
        Art Kagel -- Tuesday, 10 May 2011, at 1:17 p.m.
        • Re: Slow Query with index
          Laurie Gustin -- Tuesday, 10 May 2011, at 2:49 p.m.
          • Re: Slow Query with index
            John Miller iii -- Tuesday, 10 May 2011, at 4:17 p.m.
            • Re: Slow Query with index
              Art Kagel -- Tuesday, 10 May 2011, at 5:00 p.m.
            • Re: Slow Query with index
              Laurie Gustin -- Tuesday, 10 May 2011, at 5:13 p.m.
  • Re: Slow Query with index
    Jack Parker -- Tuesday, 10 May 2011, at 11:13 a.m.

IDS Forum is maintained by Administrator with WebBBS 5.12.