Re: [bareos-users] Questions about Virtual Full

[lst_hoe02]

Zitat von Philipp Storz <philipp.st...@bareos.com>:

Hello Andreas,

Am 25.03.2014 10:14, schrieb lst_ho...@kwsoft.de:

Zitat von lst_ho...@kwsoft.de:

Hello,


Therefore we need to know the following:


backup or does it matter at all?

we have as of now daily incremental to disk and a weekly disaster backup to tape for about 20 machines with some 2TB of data. The backup to disk does one weekly full as basis and daily incrementals for the rest of the week, the disaster tape backup runs on weekend. We now like to replace the disaster to tape backup by virtual full backups derived from the disk backups. 1.) What should we set as FileSet in the virtual? Does it need to be the same as in the original

(see sql_get.c, db_accurate_get_jobids())

So yes, you need to have the same fileset that you used before.

The virtual full backup is technically a restore of the backups that are being consolidated. Therefore the jobids that are used for the virtual full backup also are used selecting the fileset

2.) What concurrency do we get if multiple virtual fulls are started concurrently? As of now we do data spooling with all jobs running concurrently, but i suspect for virtual full to tape the jobs

will be running one by one, no?

I think you already found out in your tests...

Somewhat, yes. Without data-spool we are limited to 1. Next test will be with data-spool eg. if we can do virtual "restore" from the same disk pool concurrently??

3.) Is it possible to run another backup to disk (one pool) while the virtual fulls are still

running?

devices that could work.

As long as the volumes needed for both operations are not the same and you have two different

The source "device" is disk with multiple volumes all in the same pool, so i guess no? The source device will be blocked and not useable for backups until the virtual full has finished i suspect...

4.) Is a virtual full a fully usable independant backup or are there any special considerations

like for migration/copy jobs?

What do you mean with "special considerations like for migration/copy jobs"?

The virtual full backup is not dependent on any other backup (like a "real" full backup also)

The copy job for example is according to documentation not driectly available for restore (http://doc.bareos.org/master/html/bareos-manual-main-reference.html#x1-23600022). As there was nothing about the virtual full i decided to ask to be sure.

Any other hints and tweaks are also welcome, because documentation about virtual full is somewhat

short.

30MByte/sec to a tape drive doing 70..100MByte/sec with normal full :-(

Done some tests regarding 2. It shows that "concurrency" is 1 eg. only one job accessing the tape drive at once like expected. This is not a problem per se but the data rate is low. We got around

As we can see with "iotop" the data rate reading from disk pool is well above 100MByte so i wonder why we don't get a decent data to tape speed? Is this expected or do i have to do some tweaks?

During virtualfull, all previous backups are consolidated which means that we have to copy the blocks from the previous backups but we only write the blocks that we still need. So more data is read than written. Also, if you are writing parallel jobs to your disks, the blocks of different jobs are intermixed, and have to be sorted out during the virtual backup. If your drive is not getting its minimum write speed, you should think about configuring spooling.

Sounds sane, we will try.

Thanks

Andreas

smime.p7s
Description: S/MIME Cryptographic Signature