oracle MRP(managed recovery process) slow during backup

[saju_2]

Hi all

I am facing a problem with oracle MRP process speed. The server is a superdome with HP-UX 11.11 hosting a standby Database. We are using omniback data protector for backup.

when we fire a raw backup (oracle) the oracle recovery process associated with that DB gets very slow. Below is the output of first 5 lines of top command

CPU TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
19 ? 29966 root 154 20 35376K 5364K sleep 13:11 12.93 12.91 bma
11 ? 29965 root 148 20 35124K 2788K sleep 11:30 11.69 11.67 bma
29 ? 639 root 148 20 20368K 656K sleep 2:26 2.90 2.90 rbda
1 ? 638 root 148 20 20368K 656K sleep 2:25 2.83 2.82 rbda
25 ? 637 root 148 20 20368K 656K sleep 2:25 2.79 2.79 rbda
3 ? 4004 root -16 20 93760K 68984K run 636:58 2.78 2.78 midaemon
8 ? 635 root 148 20 20368K 656K sleep 2:27 2.76 2.75 rbda

Once the backup is over the MRP process is very fast. DBA is saying problem is purely on UNIX side. Any suggestions...????

Total no of CPU is 32
Total no of Memory is 32GB

[Steven E. Protter]

You should look at your disk layout for oracle. Restoring data is a complex write operation. That is going to take longer than a backup which is a simple, sequential write.

If your database is sitting on Raid 5 storage with the normal parity 9 configuration, every write has to be written to nine disks.

That takes time.

I find the DBA saying "UNIX is at fault" to be a bit of finger pointing. Oracle has some tools that can provide helpful information as far as where the bottlenecks actually are.

This indeed could be a Unix problem, but it could be an improperly tuned init.ora file or have other causes.

I would suggest taking a look at system performance with glance/gpm or sar commands during the backup and restore process. This will tell you where the problem is if its an OS problem.

SEP

[saju_2]

Hi Steve..

Nice to hear from you..

I haven't checked the process speed during restoration. I was checking the process speed during taking a raw backup. We take a business copy (XP 1024) of Database(Raid 0+1 H/w + 8 disk striping through LVM) to reference VG (Raid 5 H/W + 8 disk striping through LVM ). From BC we take backup.

Here 2 points are to be noted:

1. The oracle seems to be tuned properly as the MRP process is quite fast when there is no backup running.

2. If backup process (eg:bma, rdba) priority was set improperly then the MRP process will be slow for all the backup which we fire. But the MRP process works slowly only for that particular Database backup.

The backup which we take is from a reference VG which is completly different from original DB. So disk read impact will be less. Also there is a dedicated SAN switch for backup .

It seems something related to omni which gets or utlises a larger proportion of the server resources.

Any suggestions into this?