RMAN restore with TDPO statistics

[pwakim]

Hello,


I need to get some information about if anyone can give me a kind of statistics about Oracle RMAN restore throughput per one channel, when used with TDPO.

My statistics shows about 4 MB/sec per one channel restore. This number is almost the same when used with Solaris 9 or Solaris 10 client , 1GB Ethernet, UltraSPARC III+ or UltraSPARC IV client CPUs, from DISK pool or from LTO3 pool.

Is this number 4MB/sec reasonable? The result is that it needs one hour to restore about 14 GB database!. Is this normal or not?

Thank you for your help.


Regards,
Pierre

 

[hogmaster]

Ask your DBA's if the do consistency checking during backup. This can be really bad for some servers while others do really fine.
I think that you can turn that off and on per dataset without stopping th database.

My 2 cents

 

[pwakim]

Hello,

Concerning the consistency checking, I have disabled it by omitting the 'check logical' clause from the RMAN bacup scripts. Also, I disabled 'block_checking=full' from the init.ora file.

My problem is that, it takes about 20 minutes to backup a 36 GB database, while it takes about 3 hours to restore it using one single channel allocated in RMAN. The restore performance is very poor, about 4 MB per second...

I'm wondering about this restore throughput 4 MB/sec. I just need to know if anybody using TDPO has a similar problem. Also, I need to get some info from guys who have TDPO, like the throughput of TDPO during restore.


Regards,
Pierre

 

[hogmaster]

I have been running the script below to haunt down this type of problems.
In your post you say that you do the restore with a single channel, do you fdo the backup with more channels? If so can tyou try to restore with the same number of channels?

Have you tried to create a large file 10G and made a backup/restore test with that, just to make sure that you can restore a file faster? (From your post I guess that you have tried this as well).



Try this script on the tsm server, not guaranteed to work will return a divide by zero error if any session for the node you are checkn is still running. just replace ORACLE_NODE
wit the name of your tdpo client

select bytes/1073741824 as "GB" , bytes/(cast \
((end_time-start_time)seconds as integer))/1048576 as \
"MB/s" ,substr (cast((end_time)as char (10)),1,10) as "End \
date",substr (cast((end_time)as char (19)),12,8 ) as "End \
time" from summary where entity= 'ORACLE_NODE' \

 

[pwakim]

Hello,


The problem was solved after I found an Oracle Support document saying that I should add a dsm.sys parameter to the API node of Oracle database. This parameter is DISKBUFSIZE and should have the value of 32.

The stanza for Oracle node did not have the DISKBUFSIZE set at all. The default is 1024.

RMAN restore is showing tremendous results!! Restore is taking the same time it took for backup and even less.


Regards,
Pierre