Sorry for the delay.

In 10g, we can have a query that could detect differences in the number of physical reads from a table/index.
You can compare that numbers across different executions/environments, see where they start to deviate, and investigate further.
Unfortunately, that query would be valuable only if most of the wait associated with the query is IO wait.

select e.owner , e.segment_name , count(*)
from dba_hist_active_sess_history s , dba_extents e
where s.sql_id = 'fmhbn1tn0c54z'
and a.sample_time
between to_date('11/26/2012:10:00:00','MM/DD/YYYY:HH24:MI:SS')
and to_date('11/26/2012:11:00:00','MM/DD/YYYY:HH24:MI:SS')
and s.p1 = e.file_id
and s.p2 between e.block_id and (e.block_id + e.blocks - 1)
group by e.owner , e.segment_name

When I first saw your post I got a little excited at the prospect of using this sql script to debug a performance issue on a query.

However, our DB is Oracle 10g so I am curious if there is a similar way of doing the same thing for Oracle 10g as the sql_plan_line_id column (along with some others) were only introduced in 11gR1?

Many thanks in advance,

Tomás

Hinting the query to force execution order is certainly a doable solution – I’ve done it many times.

The “hinting” path has its challenges though:

First, I have to review the whole query and come with an execution order for all tables, even though there are only three tables I am interested in. As far as I know, it is not possible to specify a hint for relative execution order, i.e. TAB3 before TAB2.

Second, hinting is more difficult if the query is dynamically generated. In that case, the optimal execution order may vary depending on the dynamic clauses, so hinting the query to follow certain execution order could be risky.

Iordan Iotzov

Having said that, if the order is the case and if you still can change the code then why note hint it with just leading as it will be more stable.

select
t3.col1
from
tab3 t3 ,
tab1 t1 ,
tab2 t2
where
t3.col1 +t2.col1*0 = t1.col1
and t1.col2 in (66,166,316,416,516,616)
and t3.col1 = t2.col1
and t2.col2 in (200,300,450,550,650,750)
13 and t1.col3 = t2.col3;

Iordan Iotzov

The body of package “FORECAST” described in your presentation is available to download or is not public?

Best Regards
Norbert

Thanks for the comment.
I just posted a new entry about 11gR2 features – http://iiotzov.wordpress.com/2011/07/26/oracle-11gr2-improvements-in-managing-extended-stats-and-why-looking-for-“useful”-column-groups-is-still-relevant .

Jordan

I haven’t tried your suggested code, but just for information, Oracle has introduced with 11.2.0.2 a new feature that allows to automatically determine helpful column groups based on query workload:

http://blogs.oracle.com/optimizer/entry/how_do_i_know_what_extended_statistics_are_needed_for_a_given_workload

Randolf