Oracle 10g SQL Tuning with AWR

February 2005
Donald K. Burleson

The AWR tables contain super-useful information about the time-series execution plans for SQL statements and this repository can be used to display details about the frequency of usage for table and indexes. This article will explore these AWR tables and expose their secrets.

The AWR tables contain super-useful information about the time-series execution plans for SQL statements and this repository can be used to display details about the frequency of usage for table and indexes. This article will explore these AWR tables and expose their secrets.

We have the following AWR tables for SQL tuning.

• dba_hist_sqlstat
• dba_hist_sql_summary
• dba_hist_sql_workarea
• dba_hist_sql_plan
• dba_hist_sql_workarea_histogram

These simple tables represent a revolution in Oracle SQL tuning and we can now employ time-series techniques to optimizer SQL with better results than ever before. Let's take a closer look at these views.

dba_hist_sqlstat

This view is very similar to the v$sql view but it contains important SQL metrics for each snapshot. These include important delta (change) information on disk reads and buffer gets, as well as time-series delta information on application, I/O and concurrency wait times.

The popular WISE tool is the easiest way to analyze Oracle physical disk reads and physical disk writes (see above plot of file write I/O over time), and WISE uses STATSPACK and/or AWR data to allow you to spot hidden redo log performance trends.  WISE is our favorite Oracle tuning tool, and the only 3rd party tool that we use.
 

dba_hist_sql_plan

The dba_hist_sql_plan table contains time-series data about each object (table, index, view) involved in the query. The important columns include the cardinality, cpu_cost, io_cost and temp_space required for the object.

The query below will show the main predicates involved for each object component in a SQL execution plan:

But there is lots more information in dba_hist_sql_plan that is useful. The query below will extract importing costing information for all objects involved in each query.

Now that we see the important table structures lets examine how we can get spectacular reports from this AWR data.

Viewing table and index access with AWR

One of the problems in Oracle9i was the single bit-flag that was used to monitor index usage. You could set the flag with the "alter index xxx monitoring usage" command, and see if the index was accessed by querying the v$object_usage view.

The goal of any index access is to use the most selective index for a query, the one that produces the smallest number of rows. The Oracle data dictionary is usually quite good at this, but it is up to you to define the index. Missing function-based indexes are a common source of sub-optimal SQL execution because Oracle will not use an indexed column unless the WHERE clause matches the index column exactly.

--*************************************************
--   Copyright  2004 by Rampant TechPress Inc.
--   Free for non-commercial use!  
--   To license, e-mail info@rampant.cc
-- ************************************************

col c1 heading 'Begin|Interval|time' format a20
col c2 heading 'SQL|ID' format a13
col c3 heading 'Object|Name' format a20

col c4 heading 'Search Columns' format 999,999
col c5 heading 'Cardinality' format 999,999
col c6 heading 'Disk|Reads|Delta' format 999,999
col c7 heading 'Rows|Processed' format 999,999

break on c1 skip 2
break on c2 skip 2

select
   begin_interval_time c1,
   sql_id c2,
   object_name c3,
   search_columns c4,
   cardinality c5,
   disk_reads_delta c6,
   rows_processed_delta c7
from
   dba_hist_sql_plan
natural join
   dba_hist_snapshot
natural join
   dba_hist_sqlstat;

You can also use the dba_hist_sql_plan table to gather counts about the frequency of participation of objects inside queries.

col c1 heading 'Begin|Interval|time' format a20
col c2 heading 'SQL|ID'              format a13
col c3 heading 'Object|Name'         format a20
col c4 heading 'Object|Count'        format 999,999

break on c1 skip 2
break on c2 skip 2

select
   to_char(begin_interval_time,'yyyy-mm-dd HH24') c1,
   sql_id c2,
   object_name c3,
   count(*) c4
from
   dba_hist_sql_plan
natural join
   dba_hist_snapshot
group by
   to_char(begin_interval_time,'yyyy-mm-dd HH24'),
   sql_id,
  object_name ;

Here we can see the average SQL invocations for every database object, averaged by hour-of-the day or day-of-the-week. Understanding the SQL signature can be extremely useful for determining what objects to place in your KEEP pool, and to determining the most active tables and indexes in your database.

--*************************************************
--   Copyright 2004 by Rampant TechPress Inc.
--   Free for non-commercial use!  
--   To license, e-mail info@rampant.cc
-- ************************************************

col c1 heading 'Begin|Interval|time' format a20			

col c2 heading 'Object|Name' format a20
col c3 heading 'Search Columns' format 999,999
col c4 heading 'Disk|Reads|Delta' format 999,999
col c5 heading 'Rows|Processed' format 999,999

col c6 heading 'Access|Predicates' format a200
col c7 heading 'Filter|Predicates' format a200

break on c1 skip 2

select
   begin_interval_time c1,
   object_name c2,
   search_columns c3,
   disk_reads_delta c4,
   rows_processed_delta c5,
   access_predicates c6,
   filter_predicates c7
from
   dba_hist_sql_plan
natural join
   dba_hist_snapshot
natural join
   dba_hist_sqlstat;

The new access_predicates and filter_predicates columns are very useful because we no longer need to parse-out the WHERE clause of each SQL statement to see the access and filtering criteria for the SQL statements.

Counting object usage inside SQL

In Oracle10g we can easily see what indexes are used, when they are used and the context where they are used. Here is a simple AWR query to plot index usage:

--*************************************************
--   Copyright 2004 by Rampant TechPress Inc.
--   Free for non-commercial use!  
--   To license, e-mail info@rampant.cc
-- ************************************************

col c1 heading 'Begin|Interval|time' format a20
col c2 heading 'Search Columns' format 999,999
col c2 heading 'Invocation|Count' format a20

break on c1 skip 2

select
   begin_interval_time c1,
   count(*) c3
from
   dba_hist_sqltext
natural join
   dba_hist_snapshot
where
   lower(sql_text) like lower('%cust_name_idx%')
;

This will produce an output like this, showing a summary count of all indexes used during the snapshot interval.

Conclusion

Oracle SQL tuning is constantly different. As the data changes, Oracle must be able to accommodate the changes with new execution plans. AWR now provides complete time-series SQL execution data and further research is sure to find exciting new ways to tune SQL as the data changes over time.

For more information on SQL tuning with Oracle10g AWR, see my book "Oracle Tuning".  You can save over 30% by getting it directly from the publisher here:

http://www.rampant-books.com/book_2005_1_awr_proactive_tuning.htm

Author Bio:

Donald K. Burleson has been a DBA for more than 20 years and provides Oracle consulting for system that require high performance. The author of more than 30 books, Burleson provides Oracle consulting at www.dba-oracle.com and remote Oracle support at www.remote-dba.net.

Need an Oracle Health Check? Do you have bad performance after an upgrade?   Need to certify that your database follows best practices? BC Oracle performance gurus can quickly certify every aspect of your Oracle database and provide a complete verification that your database is fully optimized.