Bacula-users

Re: [Bacula-users] Deploying Bacula

2010-02-03 11:16:33
Subject: Re: [Bacula-users] Deploying Bacula
From: mehma sarja <mehmasarja AT gmail DOT com>
To: FredNF <fred AT nfrance DOT com>
Date: Wed, 3 Feb 2010 08:13:41 -0800
Fred,

On Wed, Feb 3, 2010 at 7:23 AM, FredNF <fred AT nfrance DOT com> wrote:
Le Wed, 03 Feb 2010 13:18:37 +0100,
Henrik Johansen <henrik AT myunix DOT dk> a écrit :

> The bulk of Bacula's DB operations are purely disk IOPS bound so I
> would argue that IOPS is way more important than RAM.
>
> > With the others advices I had, I'm planning to have a dual Xeon
> > Nehalem, with 8 ou 12 GB of RAM, and four 300 GB SAS disks in RAID
> > 10. Not sure about the OS, I'm balancing between FreeBSD and Gentoo.
> >
> > But, if someone here have a similar setup, I'm ready to hear his
> > advices and tips about my configuration.
>
> We are currently planning a large Bacula deployment (~1k machines) so
> I have been facing many of the same challenges.
>
> Regardless of whatever database you choose you'll need enough disk
> IOPS to service the DB and I don't think that 4 x 300 GB SAS are
> sufficient.
>
> A 4 disk RAID10 will give you the write IOPS equivalent to 2 disks
> and the DB is most likely going to do synchronous random writes which
> in turn is 100% disk IOPS bound.
>
> Find the tech specs of the disks you are using - they should give you
> an indication of how many random write IOPS they can handle.

I will do some bonnie++ tests :)

> Additionally, you should align your FS to the same blocksize as your
> database - 8K for postgresql if I remember correctly. It you are
> using a fixed blocksize FS where the blocksize is lower than the DB
> blocksize you could end up in a siutation where one DB operation is
> causing 2 or more disk IOPS.

Right, we never see the problem on this side. The filestem used for the
director can be:

       - if FreeBSD FFS or ZFS (ZFS is nice supported with FreeBSD 8)
       - if Gentoo or Linux Distro, ext4

In the same way, what FS do you, on the list, prefer for storage ? We
don't use tape, only disk storage. The first who talk of NTFS will need
to avoid my curses for generations.edk

>
> We backup ~35TB each week in a 3 week rotation so we just have to
> scale out in order to meet our demands and we are planning to go
> multi-DIR, multi-SD with a couple of very hefty MySQL servers to
> service them.
>
> Directors will run Linux and both our SD's and MySQL servers will run
> Solaris.

So, you plan to have dedicated databases servers, having a lightweigh
director but huge database servers ?

And, if Solaris on the SD, you'll surely use ZFS ?

> The only place where we scale up instead of our are our SD's -
> currently our 3 SD nodes have access to 300+ disks and 2 dedicated 10
> Gbit fiber links.

Wow.. Tht's really impressive. I'd like to have enough money for
building such system. But, that's not and we'll use hand-made NAS with
poor inexpensive SATA disk ;)

Right, the lowly inexpensive Coraid AOE SATA arrays are DUAL 10 GB capable with a $1K HBA. It's a simpler setup than Fiber Channel and they claim, faster. I don't work for the company but have one of their 15-bay unit.

>
> > I'm freaking out about the configs files :) They'll be really huge I
> > think.

If your filesets are standardized, which they should be else you'll go crazy, you can simplify your task. I have a directory /../.../etc/conf.d where I put filesets.conf, schedules.conf, clients.conf and the like. I can see schedules and clients being large. The documentation has some suggestions - that's where I picked up my design:

I was looking for includes. But, If I read well the documentation, I
can't specify a directory for includes. I need to give the full path
for each file ? Right ?

FileSet {
  Name = "Full Set"
  Include {
    Options {
      Compression=GZIP
      signature=SHA1
      Sparse = yes
    }
    @/etc/backup.list
  }
  Include {
     Options {
        wildfile = "*.o"
        wildfile = "*.exe"
        Exclude = yes
     }
     File = /root/myfile
     File = /usr/lib/another_file
  }
  Exclude {
     File = /tmp
  } 
} 
Mehma
------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users