śr., 12 paź 2022 o 21:37 Robert M. Candey <can...@acm.org> napisał(a):

> I've been using Bacula to back up many servers and desktops to a tape
> library since early on, but always had one server running all of the Bacula
> processes except for the individual file servers.
I assume you are familiar with architecture where every Bacula's component
can be separated from each other while maintaining a working network
connection between them.

> I'm setting up a new tape library and have new data servers, so I'm
> wondering if there is a more efficient architecture for backing up 1PB,
> mostly stored on one server and NFS-mounted to the other data servers.
> Does it make sense to run the PostgreSQL database server and storage
> servers on their own servers dedicated to Bacula?
It is always a good practice to separate different workloads for
performance reasons.

> Is there value in running the Director on one or the other?
In most cases users run all Bacula components in one server. When
performance scaling is required then Director and Catalog goes to the
separate machine and another Storage machine is added.

> Should I continue to run the storage daemon on the server that hosts the
> large data?
It is highly recommended to _NOT_ run Storage daemon on a machine that
hosts data to backup unless you backup to tape storage and you are an
expert which knows exactly what he is doing.

> I'm thinking that the NFS server might be more efficient if run on its
> own, and transfer its data over the network (100GbE) to the Bacula storage
> server attached to the tape library.  And perhaps PostgreSQL could have
> dedicated memory and CPU. I don't know what if anything is slowing down our
> backups.  Full backups take 4-6 weeks for 500 TB now.
If you backup 500TB for about 1,5 month then depending on requirements, you
should optimize your setup ASAP. You can optimize in a dozen different ways.
First (as always) you need to find your bottleneck! The point where all
your backup slows down. It could be a network, CPU, memory, database (when
not using attrib spooling and you backup a massive number of files), data
source, etc, etc. Doing optimization without backup flow analysis is a
waste of time and effort.

best regards
Radosław Korzeniewski
Bacula-users mailing list

Reply via email to