RMAN archivelog backups

[groucho47]

V 10.2.0.3 on Solaris 10

We recently went to archivelog mode for our 3.4 TB warehouse (don't ask why we did not do that earlier). We have 660 tablespaces with 702 datafiles. I inherited that mess.

Before our last full RMAN backup, I backed up the archivelogs to another storage device with the "delete input" option, so the backed up archivelogs were deleted. Oracle recommends that you back up the backed-up archivelogs to tape, but I do not have the resources to do that.

We are starved for space on the array where the archivelogs were backed up, and I want to delete the last set of backed up archivelogs on that device before backing up the current set, which will be about 300G worth of files.

The reason I ask is that when I used the delete input option when I backed up the archived logs, the sequence# for those logs still shows up in v$archived_log but not the path to them, as they no longer exist in the db_recovery_file_dest. So, I need to know that it is safe to just delete the backed up archivelogs with OS commands before backing up the current set. I have no place to test this.

Right now, we do not have the storage capacity to keep more than one complete backupset of our database, and we sometimes have to skip a week or two due to business needs. So my strategy, with one hand tied behind my back, is to make sure that I have a good, full, recent backup and available archived logs going back to a point shortly before that backup. We also have flashback database enabled.

Our Client Server group does back up our backup volumes to tape weekly, but they are just file backups with no RMAN in the mix. I cannot control that backup.

One other question if anyone knows this: I have to split the backup over two days. For the first part, I specify "duration 6:00 partial", then the next day I use the "not backed up since time 'sysdate -2'. It's clunky, but it works. My question is, can you also specify "minimize time", e.g.

duration 6:00 partial minimize time?

We have old hardware and slow connections to the backup devices, so our backup totals about 17 hours these days, not counting the archivelog backups. I'm trying to do everything I can to shrink the backup window, and hopefully learn some new tricks in the process.

Right now, I think we would do better using 100 monks with quills.

[PAVB]

archivelogs should be backup as time goes on, probably once or twice an hour - may be more often during ETL window.

[davey23uk]

a lot of people dont run archive logging on warehouses becuase it is quicker to build from the source data again rather than restore from a backup and roll forward.

[LKBrwn_DBA]

An alternate backup option is to make a full "image" copy of the database files, then schedule incremental level 1 "FOR RECOVER OF COPY..." and use block tracking to speed it up a bit.

[groucho47]

Davey.

We did not archivelog for years, but because the warehouse has grown so much and we load a lot of data, the decision was made recently to use rman, archivelogs, and flashback database for a quicker recovery. Of course, we were expected to do this with very limited resources.

Doing image backups is just not possible due to not having over 3.5 TB of space on the backup array. We flip about 300G of logs every week, so I think a block change tracking file would not really give us a quicker recovery. We drop hundreds of gigs of partions every week and add hundreds more for the next week. Every month, we do the same thing for the tables with monthly partitions. Some tables in the warehouse are truncate/refresh.

[rlondon]

Are you compressing your backups using either RMAN's "as compressed backupset option" or a third party tool?