RE: Commitment Control Overhead

["MCPARTLAND, Stan" <stanley.mcpartland@xxxxxxxxxx>]

Journaling with commitment control requires less resources that
journaling by itself.  With commitment control running, journal entries
are paged out at the commit boundary instead of when the entry is
journaled.  With the performance improvements in commitment control and
the system in general over the past few years, our standard is to use
commitment control for all transactions that span multiple records.

Regards,
Stan

--------------------------------------------------------------
Stanley A. McPartland
Bently Nevada Corporation
1617 Water St; Minden, NV  89423  USA
Voice: (702) 782-9339  Fax: (702) 782-1382
E-mail: stanley.mcpartland@bently.com
--------------------------------------------------------------

>----------
>From:  Al Barsa, Jr.[SMTP:barsa2@ibm.net]
>Sent:  Tuesday, March 10, 1998 6:16 AM
>To:    MIDRANGE-L@midrange.com
>Subject:       Re: Commitment Control Overhead
>
>At 04:56 PM 3/9/98 -0600, you wrote:
>>Can someone steer me in the direction of statistics about using Commitment
>Control.  I have only found the Advanced Backup and Recovery text talking
>about CC, so far.
>>
>>It looks like there will be at least two added disk I-Os for journaling
>the before and after images of each record; two added disk I-Os for
>journaling the start and end commitment boundary for each logical unit of
>work (LUW).  If I use OMTJRNE on my journal definition, I may avoid the
>writing of journal entries for the history of LUW resources and save on
>those I-Os.
>>
>>So, for example, if I have two write statements in my LUW, I increase the
>disk I-Os by 4 + 2, or 3 times as many disk I-Os as without CC.
>>
>First of all, commitment control requires journaling.  The overhead of
>journaling is about 5%, assuming the journal receiver is in a separate ASP
>that has not overflowed.  The new 'remote journaling' support in V4R2 is
>about 14%.
>
>Commitment control itself (after you get over the additional hurdle of
>writing and testing the programs) is relatively cheap, assuming that you
>COMMIT to the database.  In the event that you perform a ROLLBACK (in your
>own code, or uncommitted transactions are implicitly rolled back at IPL
>time), commitment control is very expensive.
>
>The design of commitment control assumes that you will ultimately COMMIT
>the transactions.  I suspect that there is one additional I-O to the
>journal indicating that the COMMIT occurred.
>
>Al
>
>
>
>
>Al Barsa, Jr. - Account for Midrange-L
>Barsa Consulting, LLC. 
>400 > 390
>
>Phone: 914-251-9400
>Fax:   914-251-9406
>
>
>
>
>
>Private mail should be sent to barsa@ibm.net
>+---
>| This is the Midrange System Mailing List!
>| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
>| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
>| Questions should be directed to the list owner/operator: david@midrange.com
>+---
>uucp
>
+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---