Home » RDBMS Server » Performance Tuning » query with high cpu usage (10.2.0.2)
query with high cpu usage [message #380878] |
Tue, 13 January 2009 14:21 ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
varunvir
Messages: 389 Registered: November 2007
|
Senior Member |
|
|
Hi Experts,
Is there any way to tune the following query using lot of CPU:-
select description,time_stamp,user_id from bhi_tracking where description like 'Multilateral:%' The explain plan for this is query is:-
---------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost |
----------------------------------------------------------------
| 0 | SELECT STATEMENT | | 178K| 6609K| 129K|
| 1 | TABLE ACCESS FULL| BHI_TRACKING | 178K| 6609K| 129K|
----------------------------------------------------------------
Bhi_tracking is used for reporting purpose and contain millions of records.Generally we keep one year data in this
table and delete the remaining.Can I drop the table
after taking export and then import it back or can i truncate
the table and then insert the rows into it to enhance
the performance.
Regards,
Varun Punj,
|
|
|
|
|
Re: query with high cpu usage [message #380984 is a reply to message #380907] |
Wed, 14 January 2009 12:31 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
varunvir
Messages: 389 Registered: November 2007
|
Senior Member |
|
|
Hi Michel,
Most of the issues are solved by gathering statistics on tables
I am wondering when Oracle automatically gather statistics
with the help of dbms_stats,then why do we need to gather stats
fo some special cases.
Can you please tell me how to find from dictionary tables if
statistics really need to be gathered on particular table
or index .
Regards,
Varun Punj,
|
|
|
Re: query with high cpu usage [message #380988 is a reply to message #380984] |
Wed, 14 January 2009 12:52 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
![](http://www.orafaq.com/forum/images/custom_avatars/102589.gif) |
Michel Cadot
Messages: 68675 Registered: March 2007 Location: Nanterre, France, http://...
|
Senior Member Account Moderator |
|
|
The following list objects you must gather statistics:
declare
l_objList dbms_stats.objectTab;
begin
dbms_stats.gather_schema_stats
( ownname => user,
options => 'LIST AUTO',
objlist => l_objList );
for i in 1..l_objList.count loop
dbms_output.put_line (l_objList(i).objName||' - '||l_objList(i).objType);
end loop;
end;
/
Regards
Michel
[Updated on: Wed, 14 January 2009 12:54] Report message to a moderator
|
|
|
|
|
|
|
|
Re: query with high cpu usage [message #551040 is a reply to message #550207] |
Sun, 15 April 2012 00:26 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
![](http://www.orafaq.com/forum/images/custom_avatars/56289.jpg) |
Kevin Meade
Messages: 2103 Registered: December 1999 Location: Connecticut USA
|
Senior Member |
|
|
select * from 100_million_row_table;
How do we optimzer the above? We don't. It is by its nature already optimized as far as the query plan goes. There comes a point at which you have an optimal query plan and you are doing the least amount of work possible. Even so, there are things we can do if we really want to:
1) use compression to reduce the size of the table and thus the I/O.
2) run PARALLEL QUERY to use more resources but reduce the elapsed time (OPTIONALLY partition the table).
Faced with the next variation:
select c1,c2,c3 from 100_million_row_table;
1) You might create an index on c1,c2,c3 in order to do a FAST FULL INDEX SCAN of the index instead of going to the table. If the index is much smaller than the table (1/3 the size or less), then you may see reduced I/O for getting the data and thus improved query times.
Then we come to:
select a.*,b.*
from
100_million_row_table a
, another_100_million_row_table b
where a.pk = b.fk
/
In addition to all the ideas above, how do we optimize this.
1) make sure your HASH JOIN has enough memory to be either OPTIMAL or ONE-PASS.
2) partition the table on the join keys so you can do PARTITION WISE HASH JOINS.
Moving forward some more, how to we deal with this:
select description,time_stamp,user_id from bhi_tracking where description like 'Multilateral:%'
1) If like 'Multilateral:%' yields less than 2% of the data then an index might be in order for fetching the rows from the table. Otherwise since we are not fetching less than 2% of the rows in the table, seriously consider using FULL TABLE SCAN.
And how would we deal with this:
select a.*,b.*
from
100_million_row_table a
, another_100_million_row_table b
where a.pk = b.fk
and a.description like 'Multilateral:%'
and b.description like '<some other string>%'
/
Again if a like expression will yeild LESS than 2% of the data from a table, then for that table an index might be a good way to fetch the data. But once data is fetched from tables we need to join.
If one of the tables yields LESS than 2% of the total rows in the other table than a NESTED LOOP JOIN between the two tables supported by an INDEX LOOKUP might be in order.
Otherwise a HASH JOIN between rows fetched from the two tables is the better choice regardless of how the data was actually fetched from the tables (FULL TABLE SCAN or INDEX ACCESS).
OK, so that is the basics I guess for today's Oracle. Remember, I said BASICS. There are always special situations. But this basic understanding will get you through 90% of your SQL tuning.
Good luck, Kevin
[Updated on: Sun, 15 April 2012 00:30] Report message to a moderator
|
|
|
Re: query with high cpu usage [message #551057 is a reply to message #551040] |
Sun, 15 April 2012 04:42 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
cookiemonster
Messages: 13937 Registered: September 2008 Location: Rainy Manchester
|
Senior Member |
|
|
@LNossov & Kevin
This thread is originally from 2009, and was resurrected recently by Jack14 with a question about Michel's code snippet.
I doubt the OP is paying attention.
|
|
|
Re: query with high cpu usage [message #551061 is a reply to message #551057] |
Sun, 15 April 2012 05:15 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
John Watson
Messages: 8944 Registered: January 2010 Location: Global Village
|
Senior Member |
|
|
This may be an old topic, but some skilled people have been contributing. My contribution would be to say "trust the optimizer", but I'ld be interested if other people agree. Take a look at this, Kevin's example of joining a pair of hundred million row tables:orcl> create table a(pk number, description varchar2(20), constraint apk primary key (pk));
Table created.
orcl> create table b(pk number, description varchar2(20), constraint bpk primary key (pk));
Table created.
orcl> exec dbms_stats.set_table_stats('scott','a',numrows=>100000000)
PL/SQL procedure successfully completed.
orcl> exec dbms_stats.set_table_stats('scott','b',numrows=>100000000)
PL/SQL procedure successfully completed.
orcl> set autotr traceonly exp
orcl> select * from a join b using (pk);
Execution Plan
----------------------------------------------------------
Plan hash value: 652036164
-------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
-------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100M| 4768M| | 351K (1)| 00:01:09 |
|* 1 | HASH JOIN | | 100M| 4768M| 3528M| 351K (1)| 00:01:09 |
| 2 | TABLE ACCESS STORAGE FULL| A | 100M| 2384M| | 552 (95)| 00:00:01 |
| 3 | TABLE ACCESS STORAGE FULL| B | 100M| 2384M| | 552 (95)| 00:00:01 |
-------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("A"."PK"="B"."PK")
orcl> A hash join, as expected. But increase the tables to a hundred billion, and what happens?orcl> exec dbms_stats.set_table_stats('scott','a',numrows=>100000000000)
PL/SQL procedure successfully completed.
orcl> exec dbms_stats.set_table_stats('scott','b',numrows=>100000000000)
PL/SQL procedure successfully completed.
orcl> select * from a join b using (pk);
Execution Plan
----------------------------------------------------------
Plan hash value: 1805968304
---------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100G| 4656G| | 1227M (1)| 66:35:51 |
| 1 | MERGE JOIN | | 100G| 4656G| | 1227M (1)| 66:35:51 |
| 2 | TABLE ACCESS BY INDEX ROWID| A | 100G| 2328G| | 826 (0)| 00:00:01 |
| 3 | INDEX FULL SCAN | APK | 100G| | | 26 (0)| 00:00:01 |
|* 4 | SORT JOIN | | 100G| 2328G| 6751G| 1227M (1)| 66:35:51 |
| 5 | TABLE ACCESS STORAGE FULL | B | 100G| 2328G| | 523K(100)| 00:01:43 |
---------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
4 - access("A"."PK"="B"."PK")
filter("A"."PK"="B"."PK")
orcl> and hinting a hash join looks ptretty disastrous:orcl> select /*+ use_hash(a b) */ * from a join b using (pk);
Execution Plan
----------------------------------------------------------
Plan hash value: 652036164
-------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time |
-------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100G| 4656G| | 54G (1)|999:59:59 |
|* 1 | HASH JOIN | | 100G| 4656G| 3445G| 54G (1)|999:59:59 |
| 2 | TABLE ACCESS STORAGE FULL| A | 100G| 2328G| | 523K(100)| 00:01:43 |
| 3 | TABLE ACCESS STORAGE FULL| B | 100G| 2328G| | 523K(100)| 00:01:43 |
-------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("A"."PK"="B"."PK")
orcl> But the clever bit comes when we let the optimizer loose with automatic parallelism:orcl> alter session set parallel_degree_policy=auto;
Session altered.
orcl> select * from a join b using (pk);
Execution Plan
----------------------------------------------------------
Plan hash value: 2461430291
------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | TQ |IN-OUT| PQ Distrib |
------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100G| 4656G| | 340M (1)| 18:29:43 | | | |
| 1 | PX COORDINATOR | | | | | | | | | |
| 2 | PX SEND QC (RANDOM) | :TQ10001 | 100G| 4656G| | 340M (1)| 18:29:43 | Q1,01 | P->S | QC (RAND) |
| 3 | MERGE JOIN | | 100G| 4656G| | 340M (1)| 18:29:43 | Q1,01 | PCWP | |
| 4 | SORT JOIN | | 100G| 2328G| | 826 (0)| 00:00:01 | Q1,01 | PCWP | |
| 5 | BUFFER SORT | | | | | | | Q1,01 | PCWC | |
| 6 | PX RECEIVE | | 100G| 2328G| | 826 (0)| 00:00:01 | Q1,01 | PCWP | |
| 7 | PX SEND BROADCAST | :TQ10000 | 100G| 2328G| | 826 (0)| 00:00:01 | | S->P | BROADCAST |
| 8 | TABLE ACCESS BY INDEX ROWID| A | 100G| 2328G| | 826 (0)| 00:00:01 | | | |
| 9 | INDEX FULL SCAN | APK | 100G| | | 26 (0)| 00:00:01 | | | |
|* 10 | SORT JOIN | | 100G| 2328G| 6751G| 340M (1)| 18:29:43 | Q1,01 | PCWP | |
| 11 | PX BLOCK ITERATOR | | 100G| 2328G| | 145K(100)| 00:00:29 | Q1,01 | PCWC | |
| 12 | TABLE ACCESS STORAGE FULL | B | 100G| 2328G| | 145K(100)| 00:00:29 | Q1,01 | PCWP | |
------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
10 - access("A"."PK"="B"."PK")
filter("A"."PK"="B"."PK")
Note
-----
- automatic DOP: Computed Degree of Parallelism is 4 because of degree limit
orcl> See how the time prediction comes down? There is a lot you can do with parallel_degree_policy, I seriously believe that it is a huge leap forward. And I love the way that I, as DBA, can enable it declaratively, without having to decorate tables or tune SQL.
|
|
|
Re: query with high cpu usage [message #551089 is a reply to message #551061] |
Sun, 15 April 2012 10:25 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) ![Go to next message Go to next message](/forum/theme/orafaq/images/down.png) |
![](http://www.orafaq.com/forum/images/custom_avatars/56289.jpg) |
Kevin Meade
Messages: 2103 Registered: December 1999 Location: Connecticut USA
|
Senior Member |
|
|
Very cool stuff to add, thanks. The extreme screnarios are always interesting examples.
In my defense I would say that though they do exist, I don't have any tables in any systems I work with that have 100 billion rows. Some day soon.
Your point is well presented though; "trust the optimizer".
Continuing in that vein, I would add that to be able to trust the optimizer, we need to give it the information it needs to do its job well. To me this means paying attention to the basics of relational databases.
1) use proper datatypes
2) use proper nullability
3) start with a normalized relational model
4) define PK/UK/FK constraints
5) collect stats with some intelligence
6) turn on costing
If you just do the basics that you are supposed to do when you design a database, 99% of all performance problems you might face will never happen in the first place. I can easily trace half the serious problems I deal with every day back to some violation of the above which kept the optimizer from doing its job.
Thanks John, I have always loved your follow ups.
As for this being an old post, sure, but someone is reading it so I don't have an issue adding to it. It is just more information on the web for people to find with a search. I have never understood why people object to adding new info to old posts.
At the expense of making my contributions sound weaker, everything can change and your mileage may vary. But the numbers I quote (like < 2% of the data (maybe next year it will be 1%)) and the strategies I use are born from my day-to-day experiences. I do in fact tune quite well using these strategies. I also believe in the three concepts:
1) simplify simplify simplify
2) 99% rule
3) bring it to the masses and it has great value
I believe Oracle is a case study simple methods that do what we need 99% of the time which can be taught to everybody. In fact, I like my summary in this thread so much I think I will add it in some form to my training class for my company.
Kevin
|
|
|
Re: query with high cpu usage [message #551196 is a reply to message #551089] |
Mon, 16 April 2012 03:48 ![Go to previous message Go to previous message](/forum/theme/orafaq/images/up.png) |
cookiemonster
Messages: 13937 Registered: September 2008 Location: Rainy Manchester
|
Senior Member |
|
|
Kevin Meade wrote on Sun, 15 April 2012 16:25As for this being an old post, sure, but someone is reading it so I don't have an issue adding to it. It is just more information on the web for people to find with a search. I have never understood why people object to adding new info to old posts.
I wasn't objecting, merely pointing out a fact that LNossov almost certainly missed since he was asking a question of the OP.
I have no issue with what you added, just warning you not to expect any feedback from the OP.
|
|
|
Goto Forum:
Current Time: Sat Sep 07 23:44:24 CDT 2024
|