Hi,

I'm currently evaluating bacula, wether It can replace my home-grown backup
software. I'm quite happy with bacula so far, yet there are still a few
issues to solve. I've configured bacula to run disk-based backups in
parallel, however, I'm not sure whether I've missed an option or I've
stumbled across one or more problems.

I'm running 1.36.3 and have "Maximum Concurrent Jobs" on the following
values:

bacula-dir.conf:
Director: 20
Storage: 15
Client: 1

bacula-sd.conf:
Storage: 20

bacula-fd.conf
FileDaemon: 20

So, every client with multiple Jobs isn't overloaded, but the backup server
itself can stream multiple clients in parallel. I have a single pool with a
single device of Type "File" configured with automatic labeling and Maximum
Volume Jobs=1, so that each volume may be recycled as soon as possible.

Now to the issues:
-If two or more jobs start at the same time, they all do write to the same
 volume, although "Maximum Volume Jobs=1" is set. 

 I assume that this option is only validated upon startup of a job and
 checks the number of already finished jobs per volume to guess wether
 the new job should write to the current or a new volume. However, this
 is likely to conflict with my idea of "recycle as soon as possible": if
 one full backup ends on such a volume together with many incremental
 jobs of other clients, any freed space cannot be reclaimed until the last
 file of this volume is expired/purged. 

 I can circumvent Bacula's current behaviour by enabling Data Spooling, 
 however, this also adds up 100% of IO load to the backup server, as 
 each client is streamed to a temporary file and afterwards copied into
 a volume.

 I would expect Bacula to count the number of currently running jobs for
 this volume when starting a new job and ask the storage daemon to use a
 new volume for the next job, if the number of already running jobs per
 volume exceeds the configured limit of jobs per volume. 

-Maybe technically related to the first issue ...
 Assume I have one client writing to volume in my single file device,
 then I add a restore job which has to retrieve files from a different
 volume, but on the same device: the restore job is queued and will
 only start after the backup job has been finished.

 This behaviour makes sens to me if I were using a tape device, but
 one of the advantages for a file device with "Random Access = yes"
 should be that random and parallel access is really available.
 Have I missed an option?

-a "true" Pool for disk-based backups: per bacula server, I have 
 multiple disks on different mountpoints, as one disk isn't enough
 to store all backups.

 Currently, clients are tied to a volume pool, which is tied to
 a single device, so I'd have to manually rearrange clients every
 now and then. No, statically changing the destination pool from
 the scheduler is far from a good solution for this issue :-)

 As far as I can see from the documentation, I'd either have to make
 each mountpoint a single device and more or less strictly attach 
 each job to an individual pool, which in turn has such a single device
 or otherwise have one device, manually create volumes and symlink
 them to the individual disks.

 My problem with the first way is, that one mountpoint will spill over,
 while a different mountpoint might have plenty of space left, so I'm
 left manually re-arranging and juggling with volumes - a work Bacula
 should do for me. The second way of solving this issue has the problem
 that I cannot use automatic labeling anymore, as any automatically
 created volumes would end up on the same soon-to-spill-over disk.

 I'd like to see a storagepool as a "true" pool: I configure multiple
 devices (each one being an individual mountpoint), add all of them
 to a single default pool and bacula literally grabs free devices 
 and volumes from it as it wishes to do so.

 Such a behaviour would also help to spread backup load to multiple
 bacula servers: just have to add more storage nodes into the
 same pool.

Am I missing a few options which make Bacula behave better in those
subjects?

Regards,

Anders


-------------------------------------------------------
This SF.Net email is sponsored by the JBoss Inc.
Get Certified Today * Register for a JBoss Training Course
Free Certification Exam for All Training Attendees Through End of 2005
Visit http://www.jboss.com/services/certification for more information
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users

Reply via email to