Query problems - Group By and Latest date

Hi all,

hopefully someone can suggest the best way of implementing the problem i am trying to resolve. We have a table which contains rows relating to tests run on our product. This table is populated from an SSIS job which parses CSV files.

There are multiple rows per serial number relating to multiple tests. The only tests i am interested in are the ones with an ID of T120. Here is the query i have so far which should make it a little easier to explain:

SELECT [SerialNumber]
,Param1
,[TimeStamp]
FROM [Build Efficiency System].[dbo].[SSIS_SCANNERDATA_TBL]
WHERE Test = 'T120'
GROUP BY SerialNumber, Param1, [TimeStamp]
ORDER BY SerialNumber

What i have above is fine to a point. The problem i am encountering is that in test T120 it specifies a part which can be be one of about 6 in field Param1. If during testing there is a problem with the part then it is replaced and the test run a second time up until the whole product passes the test. The query above returns all instances of replacements so i may have the out put as follows:

SerialNumber Param1 TimeStamp
0 Part1 15/03/07
0 Part2 15/03/07
0 Part2 16/03/07
0 Part3 15/03/03

What i really need is to only list the last part that is installed, hence the one with the latest timestamp:

SerialNumber Param1 TimeStamp

0 Part1 15/03/07

0 Part2 16/03/07

0 Part3 15/03/03

Can someone please help me to alter the above query so that it will show only those Param1 fields that have the latest date for each part.

Many thanks in advance,

Grant

This should do the trick:

SELECT [SerialNumber]
,Param1
,MAX([TimeStamp])
FROM [Build Efficiency System].[dbo].[SSIS_SCANNERDATA_TBL]
WHERE Test = 'T120'
GROUP BY SerialNumber, Param1
ORDER BY SerialNumber

You only need to take the max of your timestamp field (and remove it from the group by). The group by all fields is a bit extreme in the previous query (you could use the distinct keyword instead if you had apparent duplicate rows (i.e. replaced the part 3 times in a day).

|||Thats sorted it.
I wasn't as far of the mark in the first place as i'd thought. Thank's very much for the assistance, its much appreciated.

Cheers,

Grant|||Hi, apologies but i need one more piece of advice on this subject.

If i want to include a column with a serial number of the part that has been replaced, how would i do that. As soon as i add it, it needs to be part of an aggregate function or the group by clause. When it becomes part of the group by clause it then duplicates the part again.

Any ideas?

Thanks,

Grant|||

Is the serial number of the part in the same table - if so presumably it is different for each time that part is replaced. You can use a nested query to get that - however it will run into a problem if there are multiple records with the same date. As it stands at the moment you could not distinuish between them.

If you had records:

SerialNumber Param1 TimeStamp Part_SN
0 Part1 15/03/07 1234
0 Part2 15/03/07 1235
0 Part2 16/03/07 1236
0 Part2 16/03/07 1237
0 Part3 15/03/03 1238

How would you know which of the two Part2 items fitted on 16 Mar to give the serial number of? If there are additional fields to determine this then we need to use them

If this does not arise then the query below should serve (substitute correct fieldname for Part_SN):

SELECT B.[SerialNumber]
,B.Param1
,B.[TimeStamp]
,B.Part_SN
FROM (
SELECT [SerialNumber]
,Param1
,MAX([TimeStamp]) AS TimeStamp
FROM [Build Efficiency System].[dbo].[SSIS_SCANNERDATA_TBL]
WHERE Test = 'T120'
GROUP BY SerialNumber, Param1
) A
INNER JOIN [Build Efficiency System].[dbo].[SSIS_SCANNERDATA_TBL] B
ON (A.[SerialNumber] = B.[SerialNumber]) AND
(A.Param1 = B.Param1) AND
(A.[TimeStamp] = B.[TimeStamp]) AND
(B.Test = 'T120')
ORDER BY B.[SerialNumber]

If this is a problem then you will get multiple records in that case - one for each serial number. If the TimeStamp is a datetime which includes the time of the replacement then this will not be an issue (as long as the required serial number is the last record.

|||Thanks,

That is exactly what i wanted to do. Works like a charm.

Grantsql

query problem

The subselect in the query returns 897 rows, but when it
is included in the
where clause of an update statement, the whole table is
returned and
updated. Why? And how can I change this to only update
the 897 rows that
the subselect is returning?
UPDATE Item
SET sz_amount = 1
WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id =
dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
dbo.IRISubcategory.iri_subcategory_id =
dbo.ItemExtended.iri_subcategory_id INNER JOIN
dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
= dbo.IRICategory.iri_category_id WHERE
(dbo.Item.sz_amount = 10) AND
(dbo.IRICategory.code = '1820'))
I'm not sure but try this an tell me if it works.
UPDATE Item
SET sz_amount = 1
FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id = dbo.ItemExtended.item_id
JOIN dbo.IRISubcategory ON dbo.IRISubcategory.iri_subcategory_id =
dbo.ItemExtended.iri_subcategory_id
JOIN dbo.IRICategory ON dbo.IRISubcategory.iri_category_id =
dbo.IRICategory.iri_category_id
WHERE dbo.Item.sz_amount = 10 AND
dbo.IRICategory.code = '1820'
--Buddy
"Jamie Elliott" <jelliott@.alexlee.com> wrote in message
news:253d01c427cf$38e3e730$a501280a@.phx.gbl...
> The subselect in the query returns 897 rows, but when it
> is included in the
> where clause of an update statement, the whole table is
> returned and
> updated. Why? And how can I change this to only update
> the 897 rows that
> the subselect is returning?
>
> UPDATE Item
> SET sz_amount = 1
> WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
> JOIN dbo.ItemExtended ON dbo.Item.item_id =
> dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
> dbo.IRISubcategory.iri_subcategory_id =
> dbo.ItemExtended.iri_subcategory_id INNER JOIN
> dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
> = dbo.IRICategory.iri_category_id WHERE
> (dbo.Item.sz_amount = 10) AND
> (dbo.IRICategory.code = '1820'))
>
>
>
|||Hi Jamie,
From your descriptions, I know your subselect query will work fine and get
the correct result alone. However it goes wrong when you make it as
subselect.
Would you please have a try on Buddy Ackerman's query and tell me whether
it works. If it doesn't, would you please show me your DDL and I could
reproduce it on my machine
Thank you for your patience and cooperation. If you have any questions or
concerns, don't hesitate to let me know. We are here to be of assistance!
Sincerely yours,
Michael Cheng
Microsoft Online Support
************************************************** *********
Get Secure! - www.microsoft.com/security
This posting is provided "as is" with no warranties and confers no rights.
Please reply to newsgroups only, many thanks.

query problem

The subselect in the query returns 897 rows, but when it
is included in the
where clause of an update statement, the whole table is
returned and
updated. Why? And how can I change this to only update
the 897 rows that
the subselect is returning?
UPDATE Item
SET sz_amount = 1
WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id = dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
dbo.IRISubcategory.iri_subcategory_id = dbo.ItemExtended.iri_subcategory_id INNER JOIN
dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
= dbo.IRICategory.iri_category_id WHERE
(dbo.Item.sz_amount = 10) AND
(dbo.IRICategory.code = '1820'))I'm not sure but try this an tell me if it works.
UPDATE Item
SET sz_amount = 1
FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id = dbo.ItemExtended.item_id
JOIN dbo.IRISubcategory ON dbo.IRISubcategory.iri_subcategory_id =dbo.ItemExtended.iri_subcategory_id
JOIN dbo.IRICategory ON dbo.IRISubcategory.iri_category_id =dbo.IRICategory.iri_category_id
WHERE dbo.Item.sz_amount = 10 AND
dbo.IRICategory.code = '1820'
--Buddy
"Jamie Elliott" <jelliott@.alexlee.com> wrote in message
news:253d01c427cf$38e3e730$a501280a@.phx.gbl...
> The subselect in the query returns 897 rows, but when it
> is included in the
> where clause of an update statement, the whole table is
> returned and
> updated. Why? And how can I change this to only update
> the 897 rows that
> the subselect is returning?
>
> UPDATE Item
> SET sz_amount = 1
> WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
> JOIN dbo.ItemExtended ON dbo.Item.item_id => dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
> dbo.IRISubcategory.iri_subcategory_id => dbo.ItemExtended.iri_subcategory_id INNER JOIN
> dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
> = dbo.IRICategory.iri_category_id WHERE
> (dbo.Item.sz_amount = 10) AND
> (dbo.IRICategory.code = '1820'))
>
>
>|||Hi Jamie,
From your descriptions, I know your subselect query will work fine and get
the correct result alone. However it goes wrong when you make it as
subselect.
Would you please have a try on Buddy Ackerman's query and tell me whether
it works. If it doesn't, would you please show me your DDL and I could
reproduce it on my machine
Thank you for your patience and cooperation. If you have any questions or
concerns, don't hesitate to let me know. We are here to be of assistance!
Sincerely yours,
Michael Cheng
Microsoft Online Support
***********************************************************
Get Secure! - www.microsoft.com/security
This posting is provided "as is" with no warranties and confers no rights.
Please reply to newsgroups only, many thanks.

query problem

The subselect in the query returns 897 rows, but when it
is included in the
where clause of an update statement, the whole table is
returned and
updated. Why? And how can I change this to only update
the 897 rows that
the subselect is returning?
UPDATE Item
SET sz_amount = 1
WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id =
dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
dbo.IRISubcategory.iri_subcategory_id =
dbo.ItemExtended.iri_subcategory_id INNER JOIN
dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
= dbo.IRICategory.iri_category_id WHERE
(dbo.Item.sz_amount = 10) AND
(dbo.IRICategory.code = '1820'))I'm not sure but try this an tell me if it works.
UPDATE Item
SET sz_amount = 1
FROM dbo.Item INNER
JOIN dbo.ItemExtended ON dbo.Item.item_id = dbo.ItemExtended.item_id
JOIN dbo.IRISubcategory ON dbo.IRISubcategory.iri_subcategory_id =
dbo.ItemExtended.iri_subcategory_id
JOIN dbo.IRICategory ON dbo.IRISubcategory.iri_category_id =
dbo.IRICategory.iri_category_id
WHERE dbo.Item.sz_amount = 10 AND
dbo.IRICategory.code = '1820'
--Buddy
"Jamie Elliott" <jelliott@.alexlee.com> wrote in message
news:253d01c427cf$38e3e730$a501280a@.phx.gbl...
> The subselect in the query returns 897 rows, but when it
> is included in the
> where clause of an update statement, the whole table is
> returned and
> updated. Why? And how can I change this to only update
> the 897 rows that
> the subselect is returning?
>
> UPDATE Item
> SET sz_amount = 1
> WHERE item_id IN (SELECT Item.item_id FROM dbo.Item INNER
> JOIN dbo.ItemExtended ON dbo.Item.item_id =
> dbo.ItemExtended.item_id INNER JOIN dbo.IRISubcategory ON
> dbo.IRISubcategory.iri_subcategory_id =
> dbo.ItemExtended.iri_subcategory_id INNER JOIN
> dbo.IRICategory ON dbo.IRISubcategory.iri_category_id
> = dbo.IRICategory.iri_category_id WHERE
> (dbo.Item.sz_amount = 10) AND
> (dbo.IRICategory.code = '1820'))
>
>
>|||Hi Jamie,
From your descriptions, I know your subselect query will work fine and get
the correct result alone. However it goes wrong when you make it as
subselect.
Would you please have a try on Buddy Ackerman's query and tell me whether
it works. If it doesn't, would you please show me your DDL and I could
reproduce it on my machine
Thank you for your patience and cooperation. If you have any questions or
concerns, don't hesitate to let me know. We are here to be of assistance!
Sincerely yours,
Michael Cheng
Microsoft Online Support
****************************************
*******************
Get Secure! - www.microsoft.com/security
This posting is provided "as is" with no warranties and confers no rights.
Please reply to newsgroups only, many thanks.sql

Query Performance (am I in the ms right forum for performance Q's

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] O
N
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id])
ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] O
N
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id])
ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_idThis query involves many tables, of which you posted no information
(other than the row count). You also did not post or properly describe
the query plan. So there is not enough information to give detailed
advice.
If removing the predicate "a11.Tr_sub_type_id in (430, 433, 3530)"
increases the performance dramatically, then check out which index is
used (for that query), and add Tr_sub_type_id to this index (or create a
new index with this definition).
The query may benefit from a nonclustered index on
POt_lu_policy(Policy_id,Po_corp_unit_id)
You mention that you added an index to the TRt_lu_Trans_Subtype table,
but this seems irrelevant, since this table is not used in the query.
You could also consider running the Index Tuning Wizard.
Other generic advice:
- Make sure your statistics are up to date
- Make sure the expressions in a join clause have the same data type
definition
- Pay special attention to the clustered index definition of the largest
table
Hope this helps,
Gert-Jan
marcmc wrote:
> I get a massive bookmark in the execution plans when I run the below SQL t
o
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappea
rs
> and I get drastically improved performance. a11 is well indexed. A new ind
ex
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype]
ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id
]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other piec
es
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
> I get a massive bookmark in the execution plans when I run the below SQL t
o
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappea
rs
> and I get drastically improved performance. a11 is well indexed. A new ind
ex
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype]
ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id
]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other piec
es
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>sql

Query Performance (am I in the ms right forum for performance Q's

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_idThis query involves many tables, of which you posted no information
(other than the row count). You also did not post or properly describe
the query plan. So there is not enough information to give detailed
advice.
If removing the predicate "a11.Tr_sub_type_id in (430, 433, 3530)"
increases the performance dramatically, then check out which index is
used (for that query), and add Tr_sub_type_id to this index (or create a
new index with this definition).
The query may benefit from a nonclustered index on
POt_lu_policy(Policy_id,Po_corp_unit_id)
You mention that you added an index to the TRt_lu_Trans_Subtype table,
but this seems irrelevant, since this table is not used in the query.
You could also consider running the Index Tuning Wizard.
Other generic advice:
- Make sure your statistics are up to date
- Make sure the expressions in a join clause have the same data type
definition
- Pay special attention to the clustered index definition of the largest
table
Hope this helps,
Gert-Jan
marcmc wrote:
> I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
> and I get drastically improved performance. a11 is well indexed. A new index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
> I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
> and I get drastically improved performance. a11 is well indexed. A new index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>

Query Performance (am I in the ms right forum for performance Q's

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
This query involves many tables, of which you posted no information
(other than the row count). You also did not post or properly describe
the query plan. So there is not enough information to give detailed
advice.
If removing the predicate "a11.Tr_sub_type_id in (430, 433, 3530)"
increases the performance dramatically, then check out which index is
used (for that query), and add Tr_sub_type_id to this index (or create a
new index with this definition).
The query may benefit from a nonclustered index on
POt_lu_policy(Policy_id,Po_corp_unit_id)
You mention that you added an index to the TRt_lu_Trans_Subtype table,
but this seems irrelevant, since this table is not used in the query.
You could also consider running the Index Tuning Wizard.
Other generic advice:
- Make sure your statistics are up to date
- Make sure the expressions in a join clause have the same data type
definition
- Pay special attention to the clustered index definition of the largest
table
Hope this helps,
Gert-Jan
marcmc wrote:
> I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
> and I get drastically improved performance. a11 is well indexed. A new index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
> I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
> and I get drastically improved performance. a11 is well indexed. A new index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>

Query performance

Hi
There are two tables TableA and TableB , where TableA consists of 100,000
rows and TableB consits of 60,000 rows.
In the TableA Id is the Primary Key (Clustered Index) and in the TableB
MessageId is the Foreign Key for TableA (Id), which is a non-Clustered Index.
When i Write a query by joining two tables which will give the rows that
satisfies some condition.
Ex: SELECT TableA.Id FROM TableA
INNER JOIN TableA.Id = TableB.MessageId
WHERE TableB.ResourceId = 12
This is taking more than a second to retrieve 1200 rows to get the matching
Ids with the specified criteria.
Cannot we decrease efficiency of the query ? what are the other options so
that the query performance is decreased.
maybe you can put in the tableA some of the most requested filters.
For example, add a resourceId column in tableA, this eliminate the join.
But before this, what is the principal usage of your database?
OLTP system? (so operational) or informational.
What response time do you expect?
I think your bottleneck is your processor, the tables are small. Look at the
statistics behind your queries to find what to do or change.
"Ramnadh" <Ramnadh@.discussions.microsoft.com> a crit dans le message de
news: C8D3EF70-DD5D-4C2E-B46F-DEADC291A8FC@.microsoft.com...
> Hi
> There are two tables TableA and TableB , where TableA consists of 100,000
> rows and TableB consits of 60,000 rows.
> In the TableA Id is the Primary Key (Clustered Index) and in the TableB
> MessageId is the Foreign Key for TableA (Id), which is a non-Clustered
> Index.
> When i Write a query by joining two tables which will give the rows that
> satisfies some condition.
> Ex: SELECT TableA.Id FROM TableA
> INNER JOIN TableA.Id = TableB.MessageId
> WHERE TableB.ResourceId = 12
> This is taking more than a second to retrieve 1200 rows to get the
> matching
> Ids with the specified criteria.
> Cannot we decrease efficiency of the query ? what are the other options so
> that the query performance is decreased.
|||Ramnadh,
The example you have here is not a valid query (it's an INNER JOIN
without an ON clause). In order to speculate on why a query is slower
than you like, it's important to see the exact query. Can you cut and
paste an example that you've tested in Query Analyzer, and if possible,
include the table definitions and indexes and constraints, as well?
Steve Kass
Drew University
Ramnadh wrote:

>Hi
>There are two tables TableA and TableB , where TableA consists of 100,000
>rows and TableB consits of 60,000 rows.
>In the TableA Id is the Primary Key (Clustered Index) and in the TableB
>MessageId is the Foreign Key for TableA (Id), which is a non-Clustered Index.
>When i Write a query by joining two tables which will give the rows that
>satisfies some condition.
>Ex: SELECT TableA.Id FROM TableA
> INNER JOIN TableA.Id = TableB.MessageId
> WHERE TableB.ResourceId = 12
>This is taking more than a second to retrieve 1200 rows to get the matching
>Ids with the specified criteria.
>Cannot we decrease efficiency of the query ? what are the other options so
>that the query performance is decreased.
>

QUERY performance

when I issue a join query between T1(87 rows) and a
subselect(x10^6 rows)
the optimizer estimation row count is only 1
the query has to scan the primary clustered indexes in
all the tables that partcipate in the subselect (no other
way).
I already rum update statistics on all the objects
involved without success
that causes the server to use T1 as the outher table in a
nested loop join which makes the query run for 10 hours
instead of 10 minutes
Any ideas?
Perhaps you could try updating the statistics with the FULLSCAN option?
Peter Yeoh
http://www.yohz.com
Need smaller backups? Try MiniSQLBackup
"Amit" <amit.eshet@.intel.com> wrote in message
news:190f101c44c8e$c9932ff0$a401280a@.phx.gbl...
> when I issue a join query between T1(87 rows) and a
> subselect(x10^6 rows)
> the optimizer estimation row count is only 1
> the query has to scan the primary clustered indexes in
> all the tables that partcipate in the subselect (no other
> way).
> I already rum update statistics on all the objects
> involved without success
> that causes the server to use T1 as the outher table in a
> nested loop join which makes the query run for 10 hours
> instead of 10 minutes
> Any ideas?

Query Performance

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
Marc,
reconstruct you query, not using a11.Tr_sub_type_id in (430, 433, 3530), but
use a union (all), each with its own a11.Tr_sub_type_id = condition.
Quentin
"marcmc" <marcmc@.discussions.microsoft.com> wrote in message
news:19F88AA4-D9D6-4B1C-BAAD-A8DE2259F182@.microsoft.com...
>I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
>
> I get a massive bookmark in the execution plans when I run the below SQL
> to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>

Query Performance

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtyp
e] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id])
ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_idBookmark lookup means the index Query Optimizer uses does not contain the
columns needed in the query. I believe SQL is using the different index on
a11 table if you have "a11.Tr_sub_type_id in (430, 433, 3530)" in the WHERE
clause. One thing you can try is to add an index hint after the table name
in the FROM statement.
Is this why you posted the Index Tuning Wizard question?
"marcmc" wrote:

> I get a massive bookmark in the execution plans when I run the below SQL t
o
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappea
rs
> and I get drastically improved performance. a11 is well indexed. A new ind
ex
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subt
ype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id
]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other piec
es
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 12
0)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
>
>

query performance

2.7 Million rows in accounting_tran table
3.1 Million rows in charge table
Following query only returns 333 rows.
SELECT c.accounting_tran_id
FROM accounting_tran at
JOIN charge c on at.accounting_tran_id = c.accounting_tran_id
WHERE at.lctn_id = 'VA437'
and at.acct_tran_status_typ <> 'e'
and c.fiscal_period = 200504
Problem: query cost over 50. Can't get around table scan or index scan on
charge table. Accounting_Tran table using index s,
Charge Table Indexes:
ix_fp fiscal_period
ix_test1 accounting_tran_id, fiscal_period
ix_test2 charge_id, accounting_tran_id, fiscal_period
Any ideas on how to improve performance?
Thanks in advance,
ChrisCan you paste in here the query plan of the query to see where the
bottleneck is (if there is any)
HTH, Jens Suessmeyer.
http://www.sqlserver2005.de
--
"Chris" <Chris@.discussions.microsoft.com> schrieb im Newsbeitrag
news:F7F01F8C-5153-4A34-8195-BA3C8207CCCF@.microsoft.com...
> 2.7 Million rows in accounting_tran table
> 3.1 Million rows in charge table
> Following query only returns 333 rows.
> SELECT c.accounting_tran_id
> FROM accounting_tran at
> JOIN charge c on at.accounting_tran_id = c.accounting_tran_id
> WHERE at.lctn_id = 'VA437'
> and at.acct_tran_status_typ <> 'e'
> and c.fiscal_period = 200504
> Problem: query cost over 50. Can't get around table scan or index scan on
> charge table. Accounting_Tran table using index s,
> Charge Table Indexes:
> ix_fp fiscal_period
> ix_test1 accounting_tran_id, fiscal_period
> ix_test2 charge_id, accounting_tran_id, fiscal_period
> Any ideas on how to improve performance?
> Thanks in advance,
> Chris
>|||DDL, please, including which tables these indices are on...
"Chris" wrote:

> 2.7 Million rows in accounting_tran table
> 3.1 Million rows in charge table
> Following query only returns 333 rows.
> SELECT c.accounting_tran_id
> FROM accounting_tran at
> JOIN charge c on at.accounting_tran_id = c.accounting_tran_id
> WHERE at.lctn_id = 'VA437'
> and at.acct_tran_status_typ <> 'e'
> and c.fiscal_period = 200504
> Problem: query cost over 50. Can't get around table scan or index scan on
> charge table. Accounting_Tran table using index s,
> Charge Table Indexes:
> ix_fp fiscal_period
> ix_test1 accounting_tran_id, fiscal_period
> ix_test2 charge_id, accounting_tran_id, fiscal_period
> Any ideas on how to improve performance?
> Thanks in advance,
> Chris
>|||You can use SET SHOWPLAN option to get a text version of the query plan.
Rick Sawtell
MCT, MCSD, MCDBA
"Jens Smeyer" <Jens@.Remove_this_For_Contacting.sqlserver2005.de> wrote in
message news:Ot$0dcLVFHA.584@.TK2MSFTNGP15.phx.gbl...
> Can you paste in here the query plan of the query to see where the
> bottleneck is (if there is any)
> HTH, Jens Suessmeyer.
> --
> http://www.sqlserver2005.de
> --
> "Chris" <Chris@.discussions.microsoft.com> schrieb im Newsbeitrag
> news:F7F01F8C-5153-4A34-8195-BA3C8207CCCF@.microsoft.com...
on
>|||StmtText StmtId NodeId Parent PhysicalO
p
LogicalOp Argument DefinedValues
EstimateRows EstimateIO EstimateCPU
AvgRowSize TotalSubtreeCost OutputList Warnings Type
Parallel EstimateExecutions
-- -- -- --
-- -- --
-- -- --
-- -- -- --
-- -- -- --
SET STATISTICS PROFILE ON 76 1 0 NULL
NULL 1 NULL
NULL NULL NULL
NULL NULL NULL NULL SETSTATON
0 NULL
(1 row(s) affected)
StmtText
StmtId
NodeId Parent PhysicalOp LogicalOp
Argument
DefinedValues
EstimateRows EstimateIO
EstimateCPU AvgRowSize TotalSubtreeCost OutputList
Warnings Type
Parallel EstimateExecutions
----
----
---
-- -- -- --
--
----
---
----
-- -- --
-- --
---- --
-- -- --
SELECT c.accounting_tran_id
FROM accounting_tran at
JOIN charge c on at.accounting_tran_id = c.accounting_tran_id
WHERE at.lctn_id = 'VA437'
and at.acct_tran_status_typ <> 'e'
and c.fiscal_period = 200504 77 1 0
NULL NULL 1
NULL
7533.1636 NULL NULL
NULL 51.250423 NULL
NULL SELECT 0 NULL
|--Parallelism(Gather Streams)
77
3 1 Parallelism Gather Streams
NULL
NULL
7533.1636 0.0
4.2613383E-2 15 51.249668
[c].[accounting_tran_id] NULL PLAN_ROW
-1 1.0
|--Hash Match(Inner Join,
HASH:([at].[accounting_tran_id])=([c].[accounting_tran_id]))
77 4 3 Hash
Match Inner Join
HASH:([at].[accounting_tran_id])=([c].[accounting_tran_id])
NULL
7533.1636 0.0 0.22781526
15 51.207058 [c].[accounting_tran_id]
NULL PLAN_ROW -1
1.0
|--Bitmap(HASH:([at].[accounting_tran_id]),
DEFINE:([Bitmap1002]))
77 5 4 Bitmap
Bitmap Create HASH:([at].[accounting_tran_id])
[Bitmap1002] 6729.1655
0.0 4.5124404E-2 121 41.125393
[at].[accounting_tran_id] NULL
PLAN_ROW -1 1.0
| |--Parallelism(Repartition Streams, PARTITION
COLUMNS:([at].[accounting_tran_id]))
77 6 5 Parallelism
Repartition Streams PARTITION COLUMNS:([at].[accounting_tran_id])
NULL
6729.1655 0.0
4.5124404E-2 121 41.125393
[at].[accounting_tran_id] NULL PLAN_ROW
-1 1.0
| |--Filter(WHERE:([at].[acct_tran_status_typ]<'e' OR
[at].[acct_tran_status_typ]>'e'))
77
7 6 Filter Filter
WHERE:([at].[acct_tran_status_typ]<'e' OR
[at].[acct_tran_status_typ]>'e') NULL
6729.1655
0.0 2.9812064E-3 121 41.080269
[at].[accounting_tran_id] NULL
PLAN_ROW -1 1.0
| |--Bookmark Lookup(BOOKMARK:([Bmk1000]),
OBJECT:([RetailNightly].[dbo].[Accounting_Tran] AS [at]))
77 8 7 Bookmark Lookup Bookmark
Lookup BOOKMARK:([Bmk1000]),
OBJECT:([RetailNightly].[dbo].[Accounting_Tran] AS [at])
[at].[accounting_tran_id], [at].[acct_tran_status_typ]
6775.4692 41.056252 3.726508E-3
121 41.077286 [at].[accounting_tran_id],
[at].[acct_tran_status_typ] NULL PLAN_ROW -1
1.0
| |--Index
S(OBJECT:([RetailNightly].[dbo].[Accounting_Tran].[Lctn_Id] AS [at]),
SEEK:([at].[LCTN_ID]='VA437') ORDERED FORWARD)
77 10 8 Index
S Index S
OBJECT:([RetailNightly].[dbo].[Accounting_Tran].[Lctn_Id] AS [at]),
SEEK:([at].[LCTN_ID]='VA437') ORDERED FORWARD [Bmk1000]
6775.4692 0.01353462
3.7759214E-3 46 1.7310541E-2 [Bmk1000]
NULL PLAN_ROW
-1 1.0
|--Parallelism(Repartition Streams, PARTITION
COLUMNS:([c].[accounting_tran_id]), WHERE:(PROBE([Bitmap1002])=TRUE))
77 18 4 Parallelism
Repartition Streams PARTITION COLUMNS:([c].[accounting_tran_id]),
WHERE:(PROBE([Bitmap1002])=TRUE) NULL
74968.383 0.0
0.21370938 23 9.8538456
[c].[accounting_tran_id] NULL PLAN_ROW
-1 1.0
|--Index
Scan(OBJECT:([RetailNightly].[dbo].[Charge].[IX_test2] AS [c]),
WHERE:([c].[fiscal_period]=200504))
77 19
18 Index Scan Index Scan
OBJECT:([RetailNightly].[dbo].[Charge].[IX_test2] AS [c]),
WHERE:([c].[fiscal_period]=200504) [c].[fiscal_period],
[c].[accounting_tran_id] 74968.383 7.1294303
1.7479717 23 8.8774023
[c].[fiscal_period], [c].[accounting_tran_id] NULL PLAN_ROW
-1 1.0
(10 row(s) affected)
StmtText StmtId NodeId Parent PhysicalOp
LogicalOp Argument DefinedValues
EstimateRows EstimateIO EstimateCPU
AvgRowSize TotalSubtreeCost OutputList Warnings Type
Parallel EstimateExecutions
-- -- -- --
-- -- --
-- -- --
-- -- -- --
-- -- -- --
SET STATISTICS PROFILE OFF 78 1 0 NULL
NULL 1 NULL
NULL NULL NULL
NULL NULL NULL NULL SETSTATON
0 NULL
(1 row(s) affected)
"Jens Sü?meyer" wrote:

> Can you paste in here the query plan of the query to see where the
> bottleneck is (if there is any)
> HTH, Jens Suessmeyer.
> --
> http://www.sqlserver2005.de
> --
> "Chris" <Chris@.discussions.microsoft.com> schrieb im Newsbeitrag
> news:F7F01F8C-5153-4A34-8195-BA3C8207CCCF@.microsoft.com...
>
>|||Chris,
That is a remarkable query plan you posted. Did you actually get that
one with the indexes you described below? By the way, which index is
clustered (if any)?
I am surprised, because scanning index ix_test1 looks more favorable
than ix_test2 (assuming both are nonclustered).
The query benefits from a nonclustered index on Charge(fiscal_period,
accounting_tran_id) or a clustered index on Charge(accounting_tran_id)
You could also test if rewriting the query as below makes any
difference:
SELECT accounting_tran_id
FROM accounting_tran at
WHERE at.lctn_id = 'VA437'
AND at.acct_tran_status_typ <> 'e'
AND EXISTS (
SELECT 1
FROM charge c
WHERE c.accounting_tran_id = at.accounting_tran_id
AND c.fiscal_period = 200504
)
Hope this helps,
Gert-Jan
Chris wrote:
> 2.7 Million rows in accounting_tran table
> 3.1 Million rows in charge table
> Following query only returns 333 rows.
> SELECT c.accounting_tran_id
> FROM accounting_tran at
> JOIN charge c on at.accounting_tran_id = c.accounting_tran_id
> WHERE at.lctn_id = 'VA437'
> and at.acct_tran_status_typ <> 'e'
> and c.fiscal_period = 200504
> Problem: query cost over 50. Can't get around table scan or index scan on
> charge table. Accounting_Tran table using index s,
> Charge Table Indexes:
> ix_fp fiscal_period
> ix_test1 accounting_tran_id, fiscal_period
> ix_test2 charge_id, accounting_tran_id, fiscal_period
> Any ideas on how to improve performance?
> Thanks in advance,
> Chris

Query Performance

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_idMarc,
reconstruct you query, not using a11.Tr_sub_type_id in (430, 433, 3530), but
use a union (all), each with its own a11.Tr_sub_type_id = condition.
Quentin
"marcmc" <marcmc@.discussions.microsoft.com> wrote in message
news:19F88AA4-D9D6-4B1C-BAAD-A8DE2259F182@.microsoft.com...
>I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
>
> I get a massive bookmark in the execution plans when I run the below SQL
> to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>

Query performance

Hi
There are two tables TableA and TableB , where TableA consists of 100,000
rows and TableB consits of 60,000 rows.
In the TableA Id is the Primary Key (Clustered Index) and in the TableB
MessageId is the Foreign Key for TableA (Id), which is a non-Clustered Index.
When i Write a query by joining two tables which will give the rows that
satisfies some condition.
Ex: SELECT TableA.Id FROM TableA
INNER JOIN TableA.Id = TableB.MessageId
WHERE TableB.ResourceId = 12
This is taking more than a second to retrieve 1200 rows to get the matching
Ids with the specified criteria.
Cannot we decrease efficiency of the query ? what are the other options so
that the query performance is decreased.maybe you can put in the tableA some of the most requested filters.
For example, add a resourceId column in tableA, this eliminate the join.
But before this, what is the principal usage of your database?
OLTP system? (so operational) or informational.
What response time do you expect?
I think your bottleneck is your processor, the tables are small. Look at the
statistics behind your queries to find what to do or change.
"Ramnadh" <Ramnadh@.discussions.microsoft.com> a écrit dans le message de
news: C8D3EF70-DD5D-4C2E-B46F-DEADC291A8FC@.microsoft.com...
> Hi
> There are two tables TableA and TableB , where TableA consists of 100,000
> rows and TableB consits of 60,000 rows.
> In the TableA Id is the Primary Key (Clustered Index) and in the TableB
> MessageId is the Foreign Key for TableA (Id), which is a non-Clustered
> Index.
> When i Write a query by joining two tables which will give the rows that
> satisfies some condition.
> Ex: SELECT TableA.Id FROM TableA
> INNER JOIN TableA.Id = TableB.MessageId
> WHERE TableB.ResourceId = 12
> This is taking more than a second to retrieve 1200 rows to get the
> matching
> Ids with the specified criteria.
> Cannot we decrease efficiency of the query ? what are the other options so
> that the query performance is decreased.|||Ramnadh,
The example you have here is not a valid query (it's an INNER JOIN
without an ON clause). In order to speculate on why a query is slower
than you like, it's important to see the exact query. Can you cut and
paste an example that you've tested in Query Analyzer, and if possible,
include the table definitions and indexes and constraints, as well?
Steve Kass
Drew University
Ramnadh wrote:
>Hi
>There are two tables TableA and TableB , where TableA consists of 100,000
>rows and TableB consits of 60,000 rows.
>In the TableA Id is the Primary Key (Clustered Index) and in the TableB
>MessageId is the Foreign Key for TableA (Id), which is a non-Clustered Index.
>When i Write a query by joining two tables which will give the rows that
>satisfies some condition.
>Ex: SELECT TableA.Id FROM TableA
> INNER JOIN TableA.Id = TableB.MessageId
> WHERE TableB.ResourceId = 12
>This is taking more than a second to retrieve 1200 rows to get the matching
>Ids with the specified criteria.
>Cannot we decrease efficiency of the query ? what are the other options so
>that the query performance is decreased.
>

QUERY performance

when I issue a join query between T1(87 rows) and a
subselect(x10^6 rows)
the optimizer estimation row count is only 1
the query has to scan the primary clustered indexes in
all the tables that partcipate in the subselect (no other
way).
I already rum update statistics on all the objects
involved without success
that causes the server to use T1 as the outher table in a
nested loop join which makes the query run for 10 hours
instead of 10 minutes
Any ideas?Perhaps you could try updating the statistics with the FULLSCAN option?
Peter Yeoh
http://www.yohz.com
Need smaller backups? Try MiniSQLBackup
"Amit" <amit.eshet@.intel.com> wrote in message
news:190f101c44c8e$c9932ff0$a401280a@.phx.gbl...
> when I issue a join query between T1(87 rows) and a
> subselect(x10^6 rows)
> the optimizer estimation row count is only 1
> the query has to scan the primary clustered indexes in
> all the tables that partcipate in the subselect (no other
> way).
> I already rum update statistics on all the objects
> involved without success
> that causes the server to use T1 as the outher table in a
> nested loop join which makes the query run for 10 hours
> instead of 10 minutes
> Any ideas?

Query Performance

I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] O
N
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id])
ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_id
I get a massive bookmark in the execution plans when I run the below SQL to
return 23 rows of data...The volumens are as below. If I take out the
reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark disappears
and I get drastically improved performance. a11 is well indexed. A new index
was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype] O
N
[dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id])
ON [PRIMARY]
GO
and it has provided many other performance gains on a number of other pieces
of SQL. How do I begin to think about/code for this lack of performance.
Don't worry I don't expect you to understand the tables or business but if
any experiences have been overcome please post.
select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
select count(*) from prt_lu_product(nolock) -- Rows: 1719900
select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
select count(*) from ITv_lu_day(nolock) -- Rows: 4831
SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
-- Duration: 0:05:02.00 - 23 rows
select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
a15.ITv_year_id year_id,
count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
a11.Policy_id end)) WJXBFS1
into #ZZT5J0300BKMD00J
from fat_bse_po_risk_detail a11
join prt_lu_product a12
on (a11.Product_id = a12.Product_id)
join POt_lu_policy a13
on (a11.Policy_id = a13.Policy_id)
join vht_lu_vehicle a14
on (a11.Vehicle_id = a14.Vehicle_id)
join ITv_lu_day a15
on (a11.Inception_date_id = a15.Inception_date_id)
where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a11.Tr_sub_type_id in (430, 433, 3530)
and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
and a12.Pr_Group_id in (2, 3)
and a11.f_Ren_Flag = '4'
and a11.Po_tr_bus_type_id > 0)
group by a14.Vh_VhAll_group_id,
a15.ITv_year_idMarc,
reconstruct you query, not using a11.Tr_sub_type_id in (430, 433, 3530), but
use a union (all), each with its own a11.Tr_sub_type_id = condition.
Quentin
"marcmc" <marcmc@.discussions.microsoft.com> wrote in message
news:19F88AA4-D9D6-4B1C-BAAD-A8DE2259F182@.microsoft.com...
>I get a massive bookmark in the execution plans when I run the below SQL to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype]
ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id
]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>
>
> I get a massive bookmark in the execution plans when I run the below SQL
> to
> return 23 rows of data...The volumens are as below. If I take out the
> reference to a11.Tr_sub_type_id in (430, 433, 3530), the bookmark
> disappears
> and I get drastically improved performance. a11 is well indexed. A new
> index
> was to add the CREATE UNIQUE CLUSTERED INDEX [IX_TRt_lu_Trans_Subtype]
ON
> [dbo].[TRt_lu_Trans_Subtype]([Tr_sub_type_id], [Tr_type_id
]) ON [PRIMARY]
> GO
> and it has provided many other performance gains on a number of other
> pieces
> of SQL. How do I begin to think about/code for this lack of performance.
> Don't worry I don't expect you to understand the tables or business but if
> any experiences have been overcome please post.
>
> select count(*) from fat_bse_po_risk_detail(nolock) -- Rows: 11674571
> select count(*) from POt_lu_policy(nolock) -- Rows: 2967597
> select count(*) from prt_lu_product(nolock) -- Rows: 1719900
> select count(*) from TRt_lu_Trans_Subtype(nolock) -- Rows: 9326
> select count(*) from vht_lu_vehicle(nolock) -- Rows: 3154009
> select count(*) from ITv_lu_day(nolock) -- Rows: 4831
> SET TRANSACTION ISOLATION LEVEL READ UNCOMMITTED
> -- Duration: 0:05:02.00 - 23 rows
> select a14.Vh_VhAll_group_id Vh_VhAll_group_id,
> a15.ITv_year_id year_id,
> count(distinct(case when a12.Pr_Group_id = 5 then a11.Vehicle_id else
> a11.Policy_id end)) WJXBFS1
> into #ZZT5J0300BKMD00J
> from fat_bse_po_risk_detail a11
> join prt_lu_product a12
> on (a11.Product_id = a12.Product_id)
> join POt_lu_policy a13
> on (a11.Policy_id = a13.Policy_id)
> join vht_lu_vehicle a14
> on (a11.Vehicle_id = a14.Vehicle_id)
> join ITv_lu_day a15
> on (a11.Inception_date_id = a15.Inception_date_id)
> where (a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a11.Tr_sub_type_id in (430, 433, 3530)
> and a11.Inception_date_id >= CONVERT(datetime, '2003-04-01 00:00:00', 120)
> and a11.Inception_date_id < CONVERT(datetime, '2005-04-01 00:00:00', 120)
> and a13.Po_corp_unit_id in ('GEI', 'GNI', 'GED')
> and a12.Pr_Group_id in (2, 3)
> and a11.f_Ren_Flag = '4'
> and a11.Po_tr_bus_type_id > 0)
> group by a14.Vh_VhAll_group_id,
> a15.ITv_year_id
>