Less redo generates at the time of online backup using RMAN

Question

Hi,
can someone eleborate that why less redo generates at the time of RMAN online backup in comparison of manual hotbackup.
Thanks

Answer 1

A manual hot backup (aka "OS backup" or "scripted backup") cannot handle fractured blocks. This happens when a backup command is copying out blocks from a datafile at the same time as DBWR is updating blocks. The Oracle Block Size (8KB) is generally a multiple of the OS Block size (512Bytes to 4KB or 8KB, depending on the filesystem / raw device and how the OS command copies blocks out). It is possible for DBWR to write 8KB and the OS backup utility to be concurrently reading the same Oracle block as, say, two different 4KB reads. The first half of the Oracle block may be read by one OS call and the second half by another OS call -- the two halves would be inconsistent with each other if Oracle were writing the database block at the same time.
(Therefore a restore of such a backup would give you an inconsistent -- i.e. fractured -- block, which would be logically corrupt).
In order to handle such issues, after a BEGING BACKUP is issued, for every first change to a database block, the whole block is written to the redo stream, not just the change vectors. (If the same database block is subsequently modified while in BACKUP mode, it is only the change vectors that have to be written to the redo stream). Once the END BACKUP is issued, the redo stream does not need to capture whole database blocks.
If such a backup is restored, it still has fractured blocks. However, when applying the ArchiveLogs generated during the Backup, each such fractured block is "automagically repaired" because a whole image is restored from the ArchiveLog. (Subsequent changes are replayed as changes through the change vectors).
Hemant K Chitale
http://hemantoracledba.blogspot.com