Journaling/Commitment Control Info

["Paul Holm" <pholm@xxxxxxxxxxxxxxxxx>]

I haven't kept that close to the newest info on Journal and Commit stuff but 
given the amount of interest in this subject I asked my former coworker his 
perspective if it can help anyone.  
Thanks,  Paul Holm
 
Business: 760-432-0600   Home: 760-432-6550  
Cell:  760-415-8830
PlanetJ - Makers of WOW  (AKA... WebSphere on Steroids)



-----Original Message-----
From: Randy Johnson [mailto:randyj@xxxxxxxxxx]
Sent: Tuesday, July 06, 2004 4:22 PM
To: Kent Milligan
Cc: Adam T Stallman; KEVIN GETTING; Larry Youngren; pholm@xxxxxxxxxxxxxxxxx; 
Mark Anderson
Subject: Re: Yo.... Is Journaling/Commitment Control alive and well?


Most of my initial thoughts are similar to Kent's.  If the primary concern
is storage required for journaling, check out the MNGRCV, DLTRCV and
MINENTDTA parms of the CRTJRN and CHGJRN commands.  If it's performance,
there are scenarios where performance is enhanced by using commitment
control because database operations do not need to be forced to disk until
the transaction commits.  Using busy flags, etc. instead of the industry
standard SQL isolation levels leaves you with a non-portable, proprietary
application.  The various isolation levels provide a simple solution for
the complexity involved with implementing different levels of concurrency.

I don't have data on percentage of customers that use commitment control,
but I believe it has risen substantially over the years.  Copying Mark A in
case he has an idea.  Most applications running
on a non-AS/400 client use it since there is no such thing as COMMIT(*NONE)
on non-ISeries platforms (we support *NONE via our ODBC driver, but it is a
foreign concept to programmers on other platforms).  And obviously there
are cases (like banks, casinos, etc.) where it would be foolish to use any
algorithm to simulate commitment control that does not absolutely guarantee
the integrity of the transactions in case of a system failure, no matter
how remote the chance.

Randy Johnson
ISeries UDB Commitment Control, IBM Rochester MN
Office: 015-3/D116 Phone: (507) 253


                                                                      
You might find some good info in the Journal Performance Redbook (
http://publib-b.boulder.ibm.com/Redbooks.nsf/RedbookAbstracts/sg246286.html?Open
)