Re: Hot backup

From: HAPWLH <hapwlh_at_aol.com>
Date: 1996/02/13
Message-ID: <4frj0r$hi2_at_newsbf02.news.aol.com>#1/1


The main disadvantage to placing all tablespaces into "backup" mode before doing any of the data file backups to tape/disk is the amount of redo log information generated. When a tablespace is placed into "backup" mode, any time a modification (insert,update,delete) is made to a record, a "before image" copy of the entire BLOCK the record resides in is copied to the redo log file. Only one "before image" copy of the BLOCK is placed into the redo log file area (as opposed to a "before image" copy of the BLOCK for each record touched in the block).

As a tablespace in "backup mode" produces a significantly greater amount of redo log information (as detailed above) than when the tablespace is simply online, if a failure of the database occurs in this case, it takes a while longer to recover the data in the tablespace due to the amount of redo log information - which is also why in a lot of occurrences, the archive log files are also needed for recovery. It also takes Oracle a little while to propagate this online redo log information to a tablespace when the tablespace is placed back online after an "end backup" operation on the tablespace.

Propagating this example of a single tablespace to all of the tablespaces of a database, you can begin to get a feel for the recovery time needed for recovery if all of the tablespaces are in "backup mode" and recovery is needed.

    ***

Beyond the issue of recovery time, database performance is also greatly effected if all tablespaces are in "backup mode" at the same time. When Oracle writes to the active online redo log file, it does so sequentially.  When all of the tablespaces are in "backup mode", more data needs to be written to the redo log files, so more data is in redo allocation latch/lock contention - thus decreasing performance of the database. If you must run hot backup with all of your tablespaces in "backup mode" concurrently, I recommend monitoring the increase in this activity (redo allocation latch/lock contention) by running utlbstat immediately before the hot backup and utlestat immediately after (comparing the latch/lock contention with the data from a normal single tablespace type hot backup).

***

Hope this helps!!!

Mark Henry Received on Tue Feb 13 1996 - 00:00:00 CET

Original text of this message