> Would this work if a job is larger than the available disk storage?

If you set the disk volume size to a reasonable number (usually
something that would fit on one tape), set your number of concurrent
jobs to more than 1, and have a way of triggering a migration job
quickly, then it should work for any size backup as long as you have
enough space to store at least one full disk volume and another one (or
more) to be filling while you migrate the full one. Preallocating a set
of disk-based volumes also lets you control the maximum amount of disk
space used by a backup at any time. 

> What is the difference if I first spool the data to disk and write it
> then to tape or doing a backup to disk and then migrate it to tape?

Disk is usually your fastest medium. If the backup goes to disk-based
volumes, all the data (metadata and file data) are committed to the
database as quickly as possible and releases the client as quickly as
possible (migration is purely a server-side operation) so the director
can get on to another client. The job isn't over until both data and
metadata are completely written, so the throughput of backup jobs is
gated on the speed of your devices. 

> Where would I save time for a single job? (I've never use migration,
> please forgive me my ignorance).

You can define as many disk-based "devices" as you want (within reason),
so you can have a lot of simultaneous jobs running dumping to volumes in
the disk pool, and batch the dumps to tape w/o having to wait on clients
or have clients wait on you. 

If you truly only have one big job, it won't matter much. If you have
lots of jobs and lots of clients, the approach I'm suggesting reduces
contention for the physical tape drives and reduces wear/tear by
minimizing stop/start on the tape (you generally mount a volume, write a
LOT of data, dismount, repeat) instead of the
load/mount/write/unmount/unload/repeat cycle for multiple jobs straight
to tape. 


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bacula-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/bacula-devel

Reply via email to