Oracle Database Asynchronous Commit - Changing transactions Control

One of the most common concepts in Oracle Database is a transaction that is not completed while it definitely is not recorded in the Online Redo Logs. Once a COMMIT is executed, we can really make sure that transaction was written at Online RedoLog Files and the date are safe to use in case of Disaster.

 

Since Oracle 10gR2 this behavior can be changed using AC. (Asynchronous Commit).
Using the feature "Asynchronous Commit" it is possible to change the performance af an a DML transaction and your consistency as well. The "Asynchronous Commit" offers a higher throughput for DML transactions with many simultaneous executions. This feature improve transactions speed and can also avoid Wait events like "Log File Sync and Log File Parallel Write"

 

 

Understanding the behavior of a COMMIT

 

A transaction is usually started by an a application or user process. During the execution of a transaction, data changes are generated in buffer (memory). This memory area is defined by the parameter log_buffer. When a user or application raise a COMMIT, Oracle immediately writes the buffered data (memory) to disk (redo log files) along with the redo data to commit. Until this process is not fully completed (all data are recorded in the Online Redo Log Files) Oracle will not "release" the session.

For more information, see the article: Wait Events: "Log File Sync and Log File Parallel Write"

 

As shown in the figure above, you can change and configure the behavior of the Log Writer process, enabling Oracle Database optimize  the time of recording information in a transaction Online Redo Log Files.

 

Changing the behavior of a COMMIT

 

There are 2 options to change the COMMIT behavior:

 

1 - COMMIT STATEMENT

 

You can choose how to execute your COMMIT:
Options:
 
- IMMEDIATE WAIT
- IMMEDIATE NOWAIT
- BATCH WAIT
- BATCH NOWAIT

 

2 - Change the Session or System configuration

 

By session, just use the ALTER SESSION:
In Oracle 10gR2:
Options:
 
- IMMEDIATE WAIT
- IMMEDIATE NOWAIT
- BATCH WAIT
- BATCH NOWAIT
ALTER SESSION SET COMMIT_WRITE = '{IMMEDIATE | BATCH}, {WAIT | NOWAIT}'
 
From the Oracle 11.1 onawrds, this parameter became obsolete, and was divided into two new parameters:
- COMMIT_WAIT
- COMMIT_LOGGING
ALTER SESSION SET COMMIT_WAIT = {NOWAIT | WAIT | FORCE_WAIT}
ALTER SESSION SET COMMIT_LOGGING = '{IMMEDIATE | BATCH}'
 
By system, just use the ALTER SYSTEM:
In Oracle 10gR2:
Options:

 

- IMMEDIATE WAIT
- IMMEDIATE NOWAIT
- BATCH WAIT
- BATCH NOWAIT
ALTER SYSTEM SET COMMIT_WRITE = '{IMMEDIATE | BATCH}, {WAIT | NOWAIT}'
 
From the Oracle 11.1 onawrds, this parameter became obsolete, and was divided into two new parameters:
- COMMIT_WAIT
- COMMIT_LOGGING

 


ALTER SYSTEM SET COMMIT_WAIT = {NOWAIT | WAIT | FORCE_WAIT}
ALTER SYSTEM SET COMMIT_LOGGING = '{IMMEDIATE | BATCH}'
 
* Important considerations in Oracle 11g (R1 and / or R2)
 
- If the parameter COMMIT_WAIT was set to FORCE_WAIT, then the option (WAIT) will be used. If this parameter is set to System (ALTER SYSTEM)  or session (ALTER SESSION) then then the options via command (Transaction) will be ignored.
- The parameter COMMIT_WRITE was retained for compatibility only.
- The parameters COMMIT_WAIT and COMMIT_LOGGING take precedence over the parameter COMMIT_WRITE.
- If the parameter COMMIT_LOGGING was changed after set the parameter COMMIT_WAIT to FORCE_WAIT, then the option FORCE_WAIT will be ignored.

 

Examples

 

1 - Create control tables
create   table   commit_write_to   (  
  col_1   number,  
  col_2   varchar2 (500)  
);
create   table   commit_write_from   (  
  col_1   number,  
  col_2   varchar2 (500)  
);
 
2 - Enter Initial Values
 
begin  
for r in loop 1 .. 5000
     insert   into   commit_write_from   values   (  
r,
       dbms_random. string ('a', 500)  
);
end loop;
end;
/

 

3 - Create auxiliary tables for Events

 

create   table   event_pre    the   select   *   from   v $ session_event   where   rownum   = 0;
create   table   event_post   the   select   *   from   v $ session_event   where   rownum   = 0;

 

4 - Create procedure to execute commits

 

create   or   replace   procedure   many_commits   the  
start_time number;
end_time number;
my_sid number;
num_rows number: = 5;
begin  
     select   sid   into   my_sid  
       from   sys. v_ $ session  
      where   audsid   =   sys_context ('userenv'   'Sessionid');  
     delete   event_pre;  
     delete   event_post;  
     insert   into   event_pre   select   *   from   v $ session_event   where  sid   =   my_sid;  
     start_time   : =   DBMS_UTILITY. get_time;  
     is   r   in   (Select   *   from   commit_write_from)   loop  
         insert   into   commit_write_to   values   (  
r. col_1,
r. col_2
);
commit;
     end   loop;  
     end_time   : =   DBMS_UTILITY. get_time;  
     insert   into   event_post   select   *   from   v $ session_event   where  sid   =   my_sid;  
     DBMS_OUTPUT. new_line;  
     DBMS_OUTPUT. put_line ('time:' | | (end_time   -   start_time)   / 100 | |'seconds');  
r is in (
       select   post. event,  
             (Post. Time_waited   -   nvl (pre. time_waited, 0)) / 100time_waited,  
              post. total_waits   -   nvl (pre. total_waits, 0)        total_waits  
         from   event_pre    pre    right   join  
              event_post   post   using   (Event_id)  
        order   by   post. time_waited   -   nvl (pre. time_waited, 0)     desc 
) Loop
          num_rows   : =   num_rows   - 1;  
exit when num_rows = 0;
          DBMS_OUTPUT. put_line (  
rpad (r. event, 40) | |
to_char (r. time_waited, '9999 .99 ') | |' '| |
            to_char (r. total_waits,   '9999999 ')  
);
     end   loop;  
end   many_commits;
/

 

5 - Run inserts

 

set  Feedback   off
set   serveroutput   on   size 1000000 format   wrapped
alter   session   September   commit_write = immediate, wait;
truncate   table   commit_write_to;
exec   many_commits;

 

Results:

 

- IMMEDIATE WAIT
 
- IMMEDIATE NOWAIT
- BATCH WAIT
- BATCH NOWAIT
As a result, we can see that the biggest gain (in time) is related to COMMIT_WAIT parameter, which controls when information in a transaction are "written" in the Online RedoLog Files. Because Oracle does not wait for confirmation that the data was written in Redo logs, response time is near is 0s.

 

Benefits of using Asynchronous Commit:

 

- Oracle does not wait for confirmation that the COMMIT has been successfully completed, so the transaction can ran much more faster
- The BATCH option can be used to group several commits to a request for I / O.
- Optimized response time.

 

Observations on the use of Asynchronous Commit:

 

- The COMMIT process will return successfully even if the confirmation that the data was recorded in the Online Redo Log Files are not returned, so if the database go to a "crash" situation before the data in memory (LOG_BUFFER) being written tp Redo Log in Files, or experience any kind of IO problem, the COMMIT statement will be lost along with the data.
- The COMMIT_WRITE (Oracle 10g) and COMMIT_LOGGING/COMMIT_WAIT (Both Oracle 11g) are configured at INSTANCE level and may have different values ​​across the instances. In a RAC environment, it is recommended to keep them with the same values.
- There is no parameter to disable this feature.
- It is very important to keep in mind that a common user can change the COMMIT behavior, so as a DBA you should always guide them to the right path...

 

:D

 

 

Documentation:

 

How to Minimise Waits for 'Log File Sync'? [MOS ID 857576.1]
[]s

 

 

Victor Armbrust

Views: 1084

Tags: Asynchronous, Commit, batch, commit, commit_wait, commit_write, nowait, performance, wait

Comment

You need to be a member of Oracle Community to add comments!

Join Oracle Community

Oracle Community On

I'm not a fan of advertising, and so I will not be including any advertisements on OracleCommunity.net. However, managing this community does not come without cost! If you are willing to donate to help pay for the monthly community fees and domain services I accept Bitcoin and PayPal donations.

Donate Bitcoins

Badge

Loading…

Blog Posts

Join cases between related tables in data computing and analytics

Related computing can be: Inner join or outer join. The outer join can be further divided into left join, right join, and full join. With esProc, such kinds ofrelated computing can be easily implemented. In the discussion below, we will introduce the join operations with some examples using the table emp and table sOrder.

Table structure:

The table emp stores the employee data table, in which the…

Continue

Posted by Jim King on August 25, 2014 at 1:35am

Oracle RAC Classes

Contact me for Oracle High Availability Training / Online Classes - Dubai shans.dba@gmail.com

Posted by Shanmugam on August 18, 2014 at 1:00am

Code Examples of In-Memory Grouping for more efficient data computing and analytics

It is convenient to realize some common in-memory grouping with esProc, such as, equal grouping, alignment grouping and enumeration grouping. They are to be illustrated with the following examples.

Equal grouping 

Grouping basis of equal grouping is certain fields (or computed columns derived from fields) within a data set. Each group is a subset of original data set.

Case description: Group sales orders by the year. …

Continue

Posted by Jim King on August 15, 2014 at 12:21am

© 2014   Created by Steve Karam.

Badges  |  Report an Issue  |  Terms of Service