"Executive Summary" :
Restore speed improves greatly when increasing the number of « Backup Pieces » :
1. Backup
Tuning Backup Size : SECTION SIZE 4G
Increase Channels # : from 4 to 8
Tuning # of Datafile per Channel : maxopenfiles from 8 to 1
2. Restore
Increase Channels # : 12
Tuning number of Datafile per Channel : maxopenfiles 500
Details
A] Just a few RMAN Oracle terms to start :
Backup Set Logical structure
where the backup is stored ; includes only whole files, never partial
files.
Backup Piece A
Backup Set is composed by one or more physical binary pieces : « Backup
Piece ».
Sections Contiguous range
of blocks in a Datafile. One datafile is made by multiple sections.
SECTION SIZE Useful to
enable RMAN channels to back up a single large file in parallel. RMAN divides
the work among multiple channels, with each channel backing up one file section
in a backup piece.
MAXOPENFILES This defines the
number of files that can be read simultaneously by each channel. Default is 8.
Value of 1, datafile is read one at a time, writing the backup data into the
backup piece sequentially. Value of 8, the backup piece will contain data from every
8 datafiles mixed.
B] Now my experience with SBT_TAPE as
CommVault :
0) RMAN using CommVault out of the box
gave good Backup performance but very poor Restore speed for partial restore.
Channels = 4
Maxopenfiles = 8
For instance, allocating 4 Channels and
Maxopenfiles = 8 provide me a 240 GB / Hour throughput - thais 4 hours per TB.
Then restoring the whole database with same
Channels & MaxOpenfiles setting was ok whith a 210 GB / H rate.
The nightmare arrived when running TSPITR
or DBPITR with exclude Tablespaces clause.
The throughput went down to 32 GB per hour
........ Restoring 64 GB took 2 hours.
Backup
1 TB database rate = 240 GB / H
Restore 1 TB database rate = 210 GB / H
Partial Restore 64 GB rate =
32 GB / H !!
1) More Channels and less MaxOpenfiles
Channels = 8
Maxopenfiles = 1
Backup got a bit slower rate = 200 GB / H
Partial Restore was tiny better rate =
50 GB / H
What I understood is Performance comes
obviously with parallelism.
But allocating more Channel if they have
to scan big Backup Piece to get the data is not enough.
We need small pieces to be read by restore
process.
2a) Setting "Section Size" to
get more Backup Pieces at Backup
Channels
= 8
Maxopenfiles = 1
SECTION SIZE = 2G
2b) Setting more Open Files at Restore to
achieve efficient parallelism
Channels
= 8
Maxopenfiles = 500
Miracle has arrived :
Backup was a bit faster rate
= 260 GB / H
Partial Restore 64 GB "Rocket" rate = 480 GB / H !!
So TSPITR Restore went from 2 hours to 8
minutes !
The important point to see is the number
of Backup Pieces was 8 with my 1 TB database backuped with default setting.
It increased from 8 to 464 with the
"Section Size" setting of 2
GB.
3) A good compromise between number of
Backup Pieces and speed of Partial Restore is
Channels
= 8
Maxopenfiles = 1
SECTION SIZE = 4G
Backup rate = 260 GB / H --> 3H35
That is 232 Backup Pieces.
Channels
= 12
Maxopenfiles = 500
Total Restore rate = 605 GB / H !! --> 1H30
Partial Restore 64 GB rate = 295 GB / H --> 13 min.
C] Results
|
Tests |
Action |
Channels |
MAXOPENFILES |
MAXPIECESIZE |
SECTION
SIZE |
Size GB |
Duration |
B Pieces # |
Rate GB / H |
|
1 |
Backup |
4 |
8 |
Unlim. |
Unlim. |
950 |
04H00 |
|
238 |
|
1 |
Restore Total |
4 |
8 |
Unlim. |
Unlim. |
950 |
04H30 |
|
211 |
|
1 |
Restore Partial |
4 |
8 |
Unlim. |
Unlim. |
64 |
02H00 |
|
32 |
|
2 |
Backup |
8 |
1 |
Unlim. |
Unlim. |
910 |
04H30 |
8 |
202 |
|
2 |
Restore Partial |
8 |
1 |
Unlim. |
Unlim. |
64 |
01H15 |
5 |
51 |
|
3 |
Backup |
8 |
1 |
16 G |
Unlim. |
908 |
04H26 |
16 |
202 |
|
3 |
Restore Partial |
8 |
1 |
16 G |
Unlim. |
64 |
01H13 |
10 |
51 |
|
4 |
Backup |
8 |
1 |
Unlim. |
2 G |
908 |
03H36 |
464 |
259 |
|
4 |
Restore Partial |
8 |
1 |
Unlim. |
2 G |
64 |
8 Min. |
36 |
480 |
|
5 |
Backup |
8 |
1 |
Unlim. |
8 G |
908 |
03H36 |
103 |
259 |
|
5 |
Restore Partial |
8 ou 12 |
1 |
Unlim. |
8 G |
64 |
26 Min. |
10 |
148 |
|
6 |
Backup |
8 |
1 |
Unlim. |
4 G |
908 |
03H35 |
232 |
259 |
|
6 |
Restore Partial |
12 |
500 |
Unlim. |
4 G |
64 |
13 min. |
21 |
295 |
|
6 |
Restore Total |
12 |
500 |
Unlim. |
4 G |
908 |
1H30 |
232 |
605 |
D] RMAN scripts
Backup
#+-+-+-++-+-+-++-+-+-++-+-+-+-+-+-+-+
# RMAN backup script
# CG13 DSIT 22/03/2021 # AUC
# Standard Database 11G - 12C - 19C
# INCREMENTAL LEVEL = 0 database
# Media : CommVault
# Catalog
# FULL
#+-+-+-++-+-+-++-+-+-++-+-+-+-+-+-+-+
# 9i cmd 'backup database + archivelogs + controlfile'
# when channel type 'sbt_tape', default is "plus archivelog"
# cinematic :
# sql 'alter system archive log current' ;
# backup archivelog all
# backup database
# sql 'alter system archive log current' ;
# backup archivelog -> archivelogs generated during backup
# delete input
# autobackup of controlfile & spfile
connect target /
connect catalog IRC/<psw>@IRC_PRO
# expiration policy
configure retention policy to recovery window of 31 days ;
# expiration policy : old backups purge from CTL & RCAT
allocate channel for maintenance device type disk ;
allocate channel for maintenance device type 'sbt_tape' ;
crosscheck backup;
delete force noprompt obsolete recovery window of 31 days ;
# autobackup of controlfile & spfile after a successful backup
# in a well-known format that can be retrieved without rcat !!
# default = OFF
configure controlfile autobackup on ;
CONFIGURE CONTROLFILE AUTOBACKUP FORMAT FOR DEVICE TYPE 'sbt_tape' TO '%F';
CONFIGURE SNAPSHOT CONTROLFILE NAME TO '/work/oracle/&1/rman/snapcf_&1.f';
# avoid backup of archivelogs if ALREADY BACKUPED ONCE :
configure backup optimization on ;
# Media : 'sbt_tape'
CONFIGURE DEVICE TYPE 'sbt_tape' PARALLELISM 4 BACKUP TYPE TO COMPRESSED BACKUPSET ;
run {
# rcat/ctl & disk arclogs sync
# note : 'archive log current' is the only way of switching log in every RAC instances
# whereas 'switch logfile' is only local
resync catalog ;
Change Archivelog All Crosscheck ;
allocate channel ch1 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch2 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch3 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch4 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch5 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch6 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch7 type 'sbt_tape' maxopenfiles 1 ;
allocate channel ch8 type 'sbt_tape' maxopenfiles 1 ;
backup INCREMENTAL LEVEL = 0 database SECTION SIZE 4G format 'db_F_%d_s%s_U%U_%Y%M%D' include current controlfile tag = '&2'
plus archivelog format 'ar_F_%d_s%s_U%U_%Y%M%D' tag = '&2' ;
# RMAN Deletes Archivelog Files Not Yet Applied To Standby Database (Doc ID 1987102.1)
DELETE ARCHIVELOG ALL COMPLETED BEFORE 'sysdate-12/24' ;
# added 210915 Keep 1 Year Full Backup
change backup tag '&2' keep until time 'sysdate+365' ;
}
Tuning the Restore :
allocate
channel ch1 type 'sbt_tape'
PARMS="SBT_LIBRARY=/opt/commvault/Base64/libobk.a(shr.o), BLKSIZE=1048576
" TRACE 0;
...
allocate channel ch12 type 'sbt_tape' PARMS="SBT_LIBRARY=/opt/commvault/Base64/libobk.a(shr.o), BLKSIZE=1048576 " TRACE 0;
setlimit
channel ch1 maxopenfiles 500;
