| Oracle Server Architecture [message #59115] |
Wed, 29 October 2003 01:18  |
Daljit Singh
Messages: 290
Registered: October 2003
Location: Texas
|
Senior Member |
|
|
Hi,
My name is Daljit Singh and I m working as Oracle DBA in PATH Infotech, Delhi.
I m very much confused with one of the scenario which i m going to explain :
Suppose i update 300000 records in a table but the whole data buffers can hold only upto 200000 records at any point of time and log file can contain only 70000 records.
So after processing first 70000 records in buffer when log file get full and log switch will occur then will these uncommitted data in buffer goes to the data file or not. If not then how it can update 300000 records bcz buffer can contain maximum 200000 records and if yes then it means database will not be in consistent stage means after checkpoint too oracle will have to remove this uncommitted data from data files if instance crashes at this movement.
I am hoping a positive response from you. You are requested to help me in this case so that i can do some other work bcz my mind is totally wheeled in this scenario.
|
|
|
|
| Re: Oracle Server Architecture [message #59118 is a reply to message #59115] |
Wed, 29 October 2003 08:28  |
Thiru
Messages: 1089
Registered: May 2002
|
Senior Member |
|
|
Your SGA(ie data buffer cache,redolog buffer cache etc) NEED NOT (and in most cases CANNOT ) hold all of your transaction data. They are meant to cache data for optimial performance.
LGWR process writes the redo entries from the log buffer cache to the redolog files whenever you commit , or every 3 seconds or redolog buffer cache is 1/3rd full or when DBWR signals the LGWR to write modified buffers . So the logfiles have both committed and uncommited transaction entries. This is meant for instance recovery ( and media recovery).
Using the information contained in the redolog files rollback segments, SMON performs rollforward and rollbackward procedures during the instance recovery to make the database(datafiles) consistent and eliminate the uncommited transactions from the data file.
DBWR process optimizes I/O by performing multiblock writes to the datafiles when there are insufficient free buffers or a checkpoint or DBWR timeout etc and not necessarily during the commits.
-Thiru
|
|
|
|
| Re: Oracle Server Architecture [message #59120 is a reply to message #59118] |
Wed, 29 October 2003 21:24 |
Daljit Singh
Messages: 290
Registered: October 2003
Location: Texas
|
Senior Member |
|
|
Hi,
Thanks for pulling me out from the confusion and i also hope that in future too u will free some of ur time for me from ur busy schedule.
Thanks once again.
Bye for this time and have nice time.
|
|
|
|
] 
Blog
Search
Help
Members
Register
Login
Home
