On 04/10/10, Tim Gustafson (tjg AT soe.ucsc DOT edu) wrote:
> ...we're getting pretty pitiful throughput numbers.  When I scp a file
> from my workstation to the Bacula server, I get something like 40MB/s
> (320Mb/s).  When Bacula runs, we're lucky to get 20MB/s (160Mb/s), and
> we often get numbers closer to 10MB/s (80Mb/s).

As others have mentioned, the key is to try and work out where the
contention is.

It may be useful to run iftop on the network interfaces of the Bacula
server to see what the network IO is like, and then compare that to
iotop to see what the disk IO is like.

Bear in mind that if you are using spooling (although I assume you
aren't), the fd-client status throughput stats reported are half of the
actual native speed. This is because the throughput calculation is based
on the speed from client to destination, so the time taken is the sum of
the network transfer from the client to the spool, and then from the
spool to the tape. That, anyhow, might be a reason for the roughly 50%
factor you report.

If disk IO is the issue it might be useful to verify that your database
(what sort?) is running on a separate disk array, that your raid
controller has caching enabled (you need a BBU for this to be safe) and
that you have a good filesystem for your backup needs (the best one for
us is XFS).

Rory

-- 
Rory Campbell-Lange
rory AT campbell-lange DOT net

Campbell-Lange Workshop
www.campbell-lange.net
0207 6311 555
3 Tottenham Street London W1T 2AF
Registered in England No. 04551928

------------------------------------------------------------------------------
Virtualization is moving to the mainstream and overtaking non-virtualized
environment for deploying applications. Does it make network security 
easier or more difficult to achieve? Read this whitepaper to separate the 
two and get a better understanding.
http://p.sf.net/sfu/hp-phase2-d2d
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users