RE: 10g RAC B&R Strategies

  • From: <krish.hariharan@xxxxxxxxxxxx>
  • To: <jeffthomas24@xxxxxxxxx>, "'oracle-l'" <Oracle-L@xxxxxxxxxxxxx>
  • Date: Sun, 9 Mar 2008 20:43:07 -0600

Jeff,

 

I have seen several variants of "slow tape infrastructure". The bottom line
is that you should be able to satisfy the throughput requirements of your
backup. Having dealt with close to 80TB of backup using RMAN/NetBackup, I
would examine the assertion that you are able to scrape your backups to tape
(after a write to disk) but not able do that do that directly to tape; a
different story if you are not licensed for the VERITAS DBE (rman libobk)
then you are left to the task of scraping file systems.

 

Since you have asserted that upgrading the tape subsystem is not feasible
you can try a two pronged approach.

 

1.      Backup to /backup: I would avoid the FRA unless you have other
motivations. Implement /backup and application VIP and a listener to go with
it which RMAN can attach to and thus RMAN connection would follow the VIP
2.      Determine the bottleneck on the slow tape infrastructure. Often
times it is the network that surrounds it. In addition, depending on your
version of Veritas NetBackup you have the Disk Storage Unit (DSU) option
which can speed things up so that you are not bound by a circuit to tape.

 

Regards,

-Krish

Krish Hariharan

President/Executive Architect, Quasar Database Technologies, LLC

http://www.linkedin.com/in/quasardb

  _____  

From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Jeffery Thomas
Sent: Sunday, March 09, 2008 6:27 PM
To: oracle-l
Subject: 10g RAC B&R Strategies

 

I have some serious doubts concerning our ability to devise a simple but
robust B&R strategy 
with 10g RAC given some architectural constraints.   

We have a slow tape infrastructure where we cannot directly stream RMAN
backups to tape.
Instead, with our current 9i VERITAS RAC systems, we have a shared file
system (/backup)  to which 
we stream our RMAN backups,  with nightly O/S backups coming along and
copying the contents 
of /backup to tape.
 
We are nearly ready to start building some new 10g RAC systems on Solaris,
and the plan was to 
use a FRA.  In addition, the plan was to use ASM only, no 3rd-party
clustered file systems such as Sun 
Clusterware or VERITAS.  As a result we can have a /backup file system on
only ONE node of the
cluster, due to the lack of a clustered file system other than ASM.  

As stated in earlier threads in the Oracle-l archives as per RMAN backups to
the FRA: "FRA is primarily 
a backup to  disk to tape strategy.   Backup to the FRA then backup the FRA
to tape."

Obviously, given our current configuation, we cannot do that.    

As for backing up the FRA to disk, I found  Metalink Note 420290.1,  dated
March 2007,  "Backup Flash
Recovery Area to Disk Location":

"The 'backup recovery area' command only works with SBT (tape) channels.
There is a planned fix to allow DISK 
channels to be supported.  The RMAN disksbt library, which emulates a SBT
library (but writes backups to disk location),
can be used as a supported workaround to backup the FRA to a disk location".

So, is anyone actually doing this?   That is, backing up the FRA to disk
with the 'backup recovery area' command 
as per this Metalink note?

As I see it, our options for backups, given: 1) No direct streams to tape,
and 2) /backup file system on
one node are:

1) Create the FRA, make RMAN backups to it, and try the SBT_LIBRARY option
to backup the FRA to 
/backup.   

2) Create the FRA, make RMAN backups to it, and then backup the backupsets
to the /backup file system,
using something like this: 'backup backupset all format '/backup/...'     

3) Do not use the FRA to store RMAN backups.   Only use the FRA for archived
logs, and make our RMAN
backups to /backup, similar to our 9i strategy.   Not really a practical
option.

Of course, for all 3 options, if the node where /backup is attached goes
down, there might be a little problem.

Other than *fix the tape drives" (not going to happen), or "get a 3rd party
clusterware in addition to ASM"
(possible),  do we have other options?

Thanks,
Jeff

Other related posts: