Hello,
I wish I could get more precise details for my case. Let's consider
backup are only made on some disks.
In bacula-dir.conf, in Job resources, we may set Spool Data attribute to
yes. From the online doc, you can read: "This option should not be used
if you are writing to a disk file".
I definitly understand this advice. Why spool on disk, while the final
storage is on a disk.
However, I still want to run multiple Jobs in parallel AND avoid
interleaved backup data.
Furthermore, when I had to restore some small (< 100kb) files from a
disk volume > 50GB, I had to wait for more than 20 minutes. I don't
understand why it was so long.
Here are my questions:
1 - Is it just the OS that takes so long to set its index to the content
of the disk volume (file) or is it Bacula that treats a disk volume as a
cartridge, with sequential access instead of random access ?
2 - Based on the fact Bacula does not use random access to disk volumes,
what's the best deal between spooling data to avoid interleaved backups
data and havin interleaved backups data on disk volumes ?
3 - By the way, what do you think of havin a single ~ 60GB disk volume
or several smaller 5-10GB volumes ? Does anyone has some background
about it ?
I'm not sure I've made myself clear. Do not hesitate to ask me to
reformulate.
Arno Lehmann wrote:
> Hi,
>
> On 6/2/2006 9:00 AM, Christoph Litauer wrote:
>> Dear bacula users,
>>
>> I currently backup about 15 linux/windows servers to a LTO streamer. For
>> increased restore speed I want no parallel (interleaving) backups on the
>> tapes. So I set "Maximum Concurrent Jobs" of the tape storage to 1.
>>
>> On the other hand, collecting data should happen in parallel on all
>> clients. So I set "SpoolData = yes" and "Maximum Concurrent Jobs" in the
>> sections "Director", "FileDaemon" (in clients bacula-fd.conf) and
>> "Storage" (bacula-sd.conf) to 20.
>>
>> I expected the clients to spool in parallel but despooling sequential.
>> But the clients wait for each other doing the backups on after another.
>> Did I miss something?
>
> Not exactly, but there's another solution to your needs.
>
> Activate spooling and allow multiple concurrent jobs. In this case, the
> data for one job will be written to tape consecutively (at least the
> amount that is spooled). Despooling happens per job, so during
> despooling there's only data from one job at a time going to tape.
>
> If your spool space is smaller than what you'd need for all jobs you
> might end up with multiple chunks of data for a job, but these can be
> restored rather fast because Bacula can forward the tape between these
> chunks.
>
> This is the sort of setup I prefer because it gives me the right
> compromise between overall backup speed and restore speed.
>
> Arno
>
>
--
Baptiste MALGUY - IT Engineer SOLSOFT
PGP Fingerprint: 2A90 B6D1 855F 1584 FE68 94AC CCD6 36D0 3DC7 DCB8
www.solsoft.com - phone: +33 1 47 15 04 00 - fax: +33 1 47 15 55 09
How secure networks are managed.
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users
