RE: Journal records written to disk

Actually commitment control has very little to do with journal entries being
written, the major difference is that when under commitment control a
journal entry for a table will contain the transaction id for that
commitment cycle'.
 
If a file is journaled, any journal entries created by the DBMS are written
to disk "instantly" (sequentially). If a failure occurs you may loose the DB
records that are in mainstore but during the IPL the system will use the
journals to rebuild the missing DB changes to the table(s).
If you are under commitment control and a rollback is issued, explicitly or
implicitly (job ended) the journal entries for that transaction are used to
remove the transaction from the table(s) that were part of that transaction.
The same occurs on an IPL after a system failure, the journal entries are
used to remove the transaction from the table(s).

Duane Christen

-----Original Message-----
From: midrange-l-bounces@xxxxxxxxxxxx
[mailto:midrange-l-bounces@xxxxxxxxxxxx]On Behalf Of Raul A. Jager W.
Sent: Wednesday, June 28, 2006 3:01 PM
To: Midrange Systems Technical Discussion
Subject: Re: Journal records written to disk


Database records can (and usualy are) cached in volatile memmory, the 
journal must be written when the transactions is commited.  The OS (or 
datbamase manager) must hold the procces in the commit until the jorunal 
write is complete.
____________________________________________________________________________
___
msmith6@xxxxxxxxxxxx wrote:

has a disk controller cache ever been documented to fail - causing a
corrupted database?  Will the controller battery hold on for a Katrina
power out length of time?

-Steve
   

This is getting lower level that I was getting at.  Essentially, the 
question is if the database will wait for say, three records to get 
updated before it will acutally write the data to disk.  Is it at that 
point the OS writes the journal entries and then writes the database 
entries or are the journal entries always written directly to disk and 
the database records are written when the threshold is reached.  
Whether the cache in the controller get corrupted ro any other 
hardware function is interrupted is not what I'm getting at.

Mike.