Ram, if you connect to the database using Oracle*Net, as in rman target
sys/<passwd>@<TNS> amd allocate channel device type SBT, the backup will
actually be streamed from that client computer to the backup software.
When rman allocates channel of type SBT it maps libobk.so into its
address space. The proposition was to have this library on a single
machine, so that it can connect to multiple databases. Since that
library implements calls like sbtwrite and sbtwrite2, which are used to
send data to backup software, the machine on which that library resides
would necessarily become a communication point between all the databases
and the backup software. Thus, it would become a single point of
failure, too.
On 04/21/2015 10:43 PM, Ram Raman wrote:
Mladen or anyone else, can you explain <q>Adding the 3rd network point can severely impact the performance, since all the communication would actually go through that dedicated "backup server"</q>
I would think that the backups themselves would still run on the database servers (isn't that where the rman and oracle binaries and db files are located) and the backups would still be using the DB servers' CPU. The 'backup server' in this scenario may only have to communicate with each db server and initiate the backups on a nightly basis. Am I right in my assumption? If that is correct there will not be too much data flowing between the DB servers and the backup server; the backup server will not be a big bottleneck.
Reason I am asking is Seth's idea looks great to me, but would like to know if the backup server would be a limiting factor. Also if anyone has such a setup they can share their experiences.
Thanks
Ram.
If you like your shell scripts and are comfortable with cron,
you might be able to just enhance it enough to eliminate the
single point of failure and dramatically reduce your risks by
centralizing your backups.
Modify your rman scripts to use an Oracle wallet to
authenticate to the databases remotely through an rman client.
That way, you can take a backup without having to be on the
server and won't expose the password of a privileged account.
What about performance? NetBackup usually pushes the data from DB
server to media server. Adding the 3rd network point can severely
impact the performance, since all the communication would actually
go through that dedicated "backup server". What happens if there
are several simultaneous backups, all going through the "backup
server"? Do you need to backup all the databases at separate
times? RMAN maps libobk.so into its address space at the time when
"allocate channel device type SBT" is executed and it's libobk.so
which facilitates the communication between rman and the media
server. So, all communication for database backups would go
through this "backup server", which would not only be a
bottleneck, but also a single point of failure.
