Bacula-users

Re: [Bacula-users] (no subject)

2008-05-02 14:49:18
Subject: Re: [Bacula-users] (no subject)
From: Jean-Sébastien Hederer <hedererjs AT asperience DOT fr>
To: bacula-users AT lists.sourceforge DOT net
Date: Fri, 2 May 2008 20:48:52 +0200

Cordialement

Thomas Krwawecz III <tom AT bluegravity DOT com> a écrit :

> We're doing disk based backups over NFS to a NetApp so hardware
> compatibility is not a concern. In no way are we switching to save money. I
> just need something reliable. With Arkeia I'm lucky if scheduled backups run
> consistently over 2-4 weeks before breaking.
>
> Final questions (and I can pay if there's someone available for quick phone
> support until we get setup):
>
>
> 1) Where should I send suggestions/feature requests? To the devel list?
>
>
> 2) YES/NO: Backing up concurrent jobs/clients requires the following in
> "bacula-dir.conf", as well as the other confs, correct?
>
> Director {
>   .
>   Maximum Concurrent Jobs = 10
>   .
> }

yes.


>
>
> 3) How should pools be defined for concurrent backups? Is there a problem
> with multiple jobs writing to the same volume? Or do I need to have each job
> write to a new volume (in the same pool) with "Use Volume _Once_ = yes", or

"Use Volume _Once_ = yes" is what we're doing for our clients.


> "Maximum Volume Jobs = 1"? That doesn't seem to be efficient though. I'd
> expect some option to "write job to a unique volume, but once that job is
> done, another job can append" (to eliminate the interleaved volume blocks
> issue with simultaneous jobs writing). If we're trying to keep the total

i think  you must use spooling on disk to do that, either you make disk backup or tape backup, but I think you should better use "Use volume once"

 
> space used by a pool to 3 TB max, it doesn't seem efficient to use one job
> per volume if the volume can't fill up. Recycling by max # of volumes won't
> work (use space efficiently/keep as much data as possible) if the volumes
> aren't full. And from what I can tell there's no "max pool size" option.
>

it's not an option we use, but it can be done through use of the same as in http://www.mail-archive.com/bacula-users AT lists.sourceforge DOT net/msg29259.html:

   Maximum Volume Bytes = 5368709120
   Maximum Volumes = 46
but I didn't tried.


> Anyone follow?
>
> Here's what I have configured now:
>
> Pool {
>   Name = Weekly-Pool
>   Pool Type = Backup
>   AutoPrune = yes
>   Recycle = yes
>   Recycle Oldest Volume = yes
>   Label Format = "Weekly-Volume-"
>   Volume Retention = 14 days
>   Maximum Volumes = 48
>   Maximum Volume Bytes = 25000000000
> }
>

i think use of maximum volume and retention and recycle oldest are not safe. i would only use maximum volume and volume retention if you're sure they won't be contradictory
> Pool {
>   Name = Daily-Pool
>   Pool Type = Backup
>   AutoPrune = yes
>   Recycle = yes
>   Recycle Oldest Volume = yes
>   Label Format = "Daily-Volume-"
>   Volume Retention = 14 days
>   Maximum Volumes = 72
>   Maximum Volume Bytes = 25000000000
> }
>
>
> Thomas Krwawecz III
> --
> Blue Gravity Communications, Inc.
> 3495 Haddonfield Rd, Suite 6
> Pennsauken, NJ 08109
>
> Toll Free: 1-877-8 HOSTING
> Tel: (856) 662-9100, Fax: (856) 662-9101
> Web: http://www.bluegravity.com
>
>
> -----Original Message-----
> From: Bob Hetzel [mailto:beh AT case DOT edu]
> Sent: Thursday, May 01, 2008 10:38 AM
> To: Thomas Krwawecz III; Bacula-users AT lists.sourceforge DOT net
> Subject: Re: [Bacula-users] Migration from Arkeia, Other Questions
>
> Thomas,
>
> As others have suggested, you'll should probably use spooling, unless
> you're doing disk based backups. But if your spool directory is located
> on a busy disk (such as where the database lives, or temp stuff for the
> database) you may not help things.  In my case, to get spooling to work
> well I needed to create a raid-0 stripe set both to get enough clients
> spooling at the same time and to keep pace with my two LTO-2 tape
> drives.  In my case my raid-0 set is four 146 GB SAS 10,000 rpm drives.
>   I can do 8 concurrent backups pretty well.  The clients are all
> desktops and laptops so they can't go anywhere near as fast as my tape
> drive, hence another reason to do spooling.
>
> The speed you get will depend heavily on the hardware involved.  If the
> clients have slow CPU or disk drives, filesystems with a million small
> files (i.e. a mail or web server) will go very slow.  Check the CPU load
>   on the clients being backed up. If they're pegged at 100%, then
> perhaps that's your bottleneck.  Likewise with the bacula server.  As I
> type this e-mail, my two socket dual core xeon 5160 backup server has
> load (from the uptime or top command) running between 9 and 10 which is
> rather high but since it's dedicated to backups that's ok.
>
> Bacula's handling of many small files on my web server has shown an
> improvement over Veritas Backup Exec (v. 9.1) for me, but Backup Exec
> doesn't do spooling or concurrency in that version.  However I believe I
> saw an improvement even without spooling/concurrency turned on.
>
> In short, pretty much any solution is going to cost some money to do
> well.  Even if you avoid the high license fees by going open source
> you'll find that the superior algorithms in bacula can't get around
> inadequacies of all the hardware involved (if any).  Unfortunately, that
> means that to compare performance of bacula with other systems you have
> to fully implement it.
>
>> From: "Thomas Krwawecz III" <tom AT bluegravity DOT com>
>> Subject: [Bacula-users] Migration from Arkeia, Other Questions
>> To: <Bacula-users AT lists.sourceforge DOT net>
>> Message-ID: <F3378DED79D440428FD8F90156DA39F7@TLK3HOME>
>> Content-Type: text/plain; format=flowed; charset="Windows-1252";
>>         reply-type=original
>>
>> I'm wondering if anyone has migrated from Arkeia, and how much of a
>> performance increase was realized. I'm setting things up now and so far it
>
>> seems to be much a better product. Fortunately the .ignoredir patch was
>> released earlier this month or that would have been a deal breaker. We
>> exclude hundreds of dirs from backup using this method.
>>
>> We're backing up 50 (out of 200 potential) servers, 580 GB for a full
>> backup, on a GigE local network. A few servers have 1-2 million small
> files
>> so we're looking for suggestions on how to speed up backups on these
>> machines. With Arkeia, they're taking 8 hours to finish. Feedback
>> appreciated.
>>
>> One thing I'm having a hard time understanding. Can jobs run concurrently
>> (backup 25 servers at a time) to the same volume? Or is it suggested that
>> each job/client backup to its own volume?
>>
>> Thanks in advance for any assistance. I'd love to contribute suggestions,
>> help, money, and/or hardware if this works out. These yearly "maintenance"
>
>> fees are getting crazy for completely unreliable software. Not to mention
>> most companies are charging more money for larger VTL sizes - which I find
>
>> ridiculous.
>>
>> Thomas Krwawecz III
>> --
>> Blue Gravity Communications, Inc.
>> 3495 Haddonfield Rd, Suite 6
>> Pennsauken, NJ 08109
>>
>> Toll Free: 1-877-8 HOSTING
>> Tel: (856) 662-9100, Fax: (856) 662-9101
>> Web: http://www.bluegravity.com
>>
>>
>>
>
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
> Don't miss this year's exciting event. There's still time to save $100.
> Use priority code J8TL2D2.
> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
> _______________________________________________
> Bacula-users mailing list
> Bacula-users AT lists.sourceforge DOT net
> https://lists.sourceforge.net/lists/listinfo/bacula-users
>

Jean-Sébastien Hederer - Gérant ASPerience
Conseil, intégration, hébergement de systèmes informatiques spécialisé en
solutions logiciel libre de gestion d'entreprise
Mob.: 06 69 56 21 49
Tél.: 02 51 13 50 94
Fax.: 02 40 68 70 06
XMPP/Jabber: hedererjs AT jabbernet DOT eu
http://www.asperience.fr
http://www.viadeo.com/fr/profile/jean-sebastien.hederer


Attachment: pgpMkEujV4PDm.pgp
Description: Signature =?utf-8?b?bnVtw6lyaXF1ZQ==?= PGP

Attachment: bindUgRLbAMeB.bin
Description: Clef publique PGP

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users