Hi
How about archiving the contents of logbuffer directly to archived logs,
instead of writing it to the intermediary store called online redologs and then archiving it to archived redo logs.
Printable View
Hi
How about archiving the contents of logbuffer directly to archived logs,
instead of writing it to the intermediary store called online redologs and then archiving it to archived redo logs.
my 2 cents ...
as par Oracle architecture, log buffer size should be 4MB or few more MBs more. In this case, there will be hundreds of archive logs will generate every hour which is very I/O expensive.
as we don't have groups in log buffer, transactions has to wait till the current log buffer archived which leads to system hungs.
as log buffer is in memory, we will loose data if instance crashed. no roll forward information will be available to do crash recovery.
stand by databases which are in high available and high protection mode need to find other methods.
replication software like Oracle Golden Gate should be no longer real time because mining from archive log and apply SQL takes time.
Just curious to know, how did you got this idea which needs to change Oracle architecture? did you see any advantages?
Thanks,
Hi
This was a interview question.
I wanted to hear from the experts my answer was incidentally similar to yours as oracle will batch writes and write to the logs and if there are no online logs then 100's of archived logs needs to be created depending on the transaction volume and performance maybe a hit
I dont however agree with this statement
as log buffer is in memory, we will loose data if instance crashed. no roll forward information will be available to do crash recovery.
as the log buffer contains only changed data not commited ones
Quote:
I dont however agree with this statement
as log buffer is in memory, we will loose data if instance crashed. no roll forward information will be available to do crash recovery.
as the log buffer contains only changed data not committed ones
agree with you hrishy. However, if there is no redo log concept, apparently Oracle has to maintain the committed records also in log buffer until flushed to archive log... :)
Thanks,