Understanding arcredo backup logic: reducing tape mounts

[jim_dalton]

Hello all

I have several oracle instances running under Solaris. I back up using OIP.

One feature that slows its normal rapid progress is arcredo handling.

I'm going to tape: it bundles up arcredos and writes them in groups to tape (50M redos bundled up into 2.5Gb written as a single file to tape - not sure how that works) , then it creates another backup job and repeats. It takes seconds to write to tape but as these are separate jobs, it has to go through tape mount logic each time, so these jobs take minutes when you factor in the tape movement.

If anyone can clarify how this can be make more effiient by reducing tape mounts I'd be most appreciative. Obviously going to disk would be one way, but we have what we have, for now at least.

I'm running with default set of params in OIP: I find documentation around OIP settings a bit scarce.

Regards,Jim  

[RiaanBadenhorst · Accepted solution]

Have you tried adjusting these?

Maximum Backup Set Limits parameters:
■ Number of files per backup set (FILESPERSET) specifies the maximum number of input files to include in each output backup set.
■ Size of the backup set (KB) (MAXSETSIZE) specifies a maximum size for a backup set in kilobytes.

[Marianne]

You may want to consider increasing MEDIA_UNMOUNT_DELAY on the master. 

The default is 180 (3 min).

Another option is to implement DSSU for Archive Log backups. 
Backup to disk for these type of backups is so much more efficient.
Stage to tape once every hour or couple of times per day.

[jim_dalton]

Those are the params I refer to as being a bit scarce as to meaning Riaan. No mention of defaults, but seems like greyed out 0, meaning unset. But as it says maximum limits, then you would imagine that no setting means...no limit? Ive got a feeling it could be rman thats coming into play here, following that logic.

Jim

[RiaanBadenhorst]

Yes, those are RMAN parameters which you can set by ticking the box and specifying a value. If not specified it would take the RMAN defaults. You should be able to check those by logging into the client, and rman, and then typing show all.

For Oracle + dedupe its recommended to set FILESPERSET to 1, so that you get the same dbf file each time to compare (dedupe) against. So I think if you increase your FILESPERSET to 100 you should get 5GB of logs.

Don't have enough Oracle (logs) to test that for you :)

[jim_dalton]

Confirmed that param does do what we thought ie at the end of the day it can be used to stop the dismount by increasing the files per set to something relevant. It cant be changed in rman, but if you override it via the oracle tab in OIP it shows up in the backup job saying FILESPERSET overridden.

I should add that rman works out this setting and so the default is not unlimited.

Jim

[RiaanBadenhorst]

Thanks for the update.