alter system set "_BACKUP_KSFQ_BUFSZ"=1048576;

RMAN Restore Performance from Tape is Very Poor [ID 850988.1]
		Modified 30-JUN-2009     Type PROBLEM     Status MODERATED

In this Document
  Symptoms
  Cause
  Solution
  References

---

This document is being delivered to you via Oracle Support's Rapid Visibility (RaV) process, and therefore has not been subject to an independent technical review.

Applies to:

Oracle Server - Enterprise Edition - Version: 10.2.0.4 to 11.1.0.7
This problem can occur on any platform.

Symptoms

RMAN restore from sbt backup is very slow compared to backup speed.
Backup rate is 60-90Mb/sec.
Restore rate is 6Mb/sec.
Backup is hardware multiplexed but media manager does support multiplexed backups.
Restore validate runtime is VERY fast so problem is the write back to disk.

Cause

RMAN is using a 128Kb buffer size for the datafiles when restoring back to disk.
The input buffer size used during backup is 1Mb.
To confirm that you are hitting the same problem, query v$backup_async IO and check buffer_size for the backup eg

SQL>select device_type, type, buffer_size from v$backup_async_io where
open_time > to_date('30-jun-09 15:14:00', 'dd-mon-rr hh24:mi:ss') and
close_time < to_date('30-jun-09 16:14:00', 'dd-mon-rr hh24:mi:ss') ;

DEVICE_TYPE TYPE BUFFER_SIZE
----------------- --------- -----------
DISK AGGREGATE 0
DISK                 INPUT 1048576
...etc
SBT_TAPE            OUTPUT  262144
Buffer size used for read from disk is 1Mb.
Then query the restore:

SQL> select device_type, type, buffer_size from v$backup_async_io where
open_time > to_date('30-jun-09 14:50:00', 'dd-mon-rr hh24:mi:ss') and
close_time < to_date('30-jun-09 14:52:00', 'dd-mon-rr hh24:mi:ss') ;

DEVICE_TYPE TYPE BUFFER_SIZE
----------------- --------- -----------
DISK     AGGREGATE 0
SBT_TAPE        INPUT 262144
DISK              OUTPUT 131072
DISK              OUTPUT 131072
DISK              OUTPUT 131072
DISK              OUTPUT 131072

Buffer size used for write back to disk is 128Kb

Solution

Change the buffer size by setting the following hidden parameter:
SQL>alter system set "_BACKUP_KSFQ_BUFSZ"=1048576;
Users restoring to ASM may want to set this to the ASM stripe size.

Problem can affect any release prior to 11.2.
Please note:
Setting _BACKUP_KSFQ_BUFSZ=1048576 will benefit restore performance overall where you can see that a buffer size of 128Kb is being used.   However,  if restore validate is also taking a long time then setting this hidden parameter alone  may not be enough, particularly if you have hardware multiplexed the original backup (allocated > 1 channel per physical tape device). If this is the case then refer to Note 740911.1: RMAN Restore Performance.