Recommendation for backup target - Linux or FreeNAS?

[rfn]

Hi,

I'm going to build a box to be a backup target for Veeam. Would I get the best performance by installing a Linux like CentOS and use it as a Linux target directly in Veeam or by installing FreeNAS and using a CIFS or iSCSI share?

I will use it to create a pool of at least two EONstor DAS connected with Fibre Channel. I have tested it and it works by using LVM and XFS in CentOS but I'm worried about what happens if I loose the Linux box. Will I be able to create a new one and access the data or will it be very difficult? I imagine that it's possible to save the config of a dedicated storage solution like FreeNAS and that it will be easier to rebuild the box.

Regards,
René Frej Nielsen

[tsightler]

It's really a pretty tough call. I personally would suggest using the Linux box, but you can always use OpenFiler and get both an appliance and Linux system in one.

If you were to decide to go with CentOS it's pretty easy to backup the LVM config but even without it it is typically not difficult to recover. I'm assuming the OS will be installed on it's own dedicated drives?

[rfn]

Hi,

I'm actually already using a CentOS VM as a target. It's running on a production host but with a dedicated FC HBA (using pass-through) that's connected to backup storage in another server room (there's obviosly a fiber connecting the two rooms ). It works great but there's two problems:

1. I only started reading up on LVM a few day ago and know not much about it. I have read some warnings about the dangers with LVM and that it can be difficult to rescue data if something goes wrong. If it's pretty easy to backup the config so that I can create a new CentOS box then I might just do that since this setup is working.

2. I need the Linux VM to run in the secondary server room or create a physical machine with Linux on it. If we were to loose our primary server room because of a disaster (tomorrow) then I would loose the Linux VM that's connecting Veeam to the backup storage and I would also loose the HBA.

I have two older vSphere hosts in the secondary server room but they don't support pass-through so my only option is to install CentOS (or some NAS solution) on a dedicated physical machine. The question was what this machine should be running and your reply is very appreciated. I also guess that hardware support is somewhat better in Linux compared to FreeBSD?

BTW, I need to pool the two EONstor DAS' because I have one VM with 25 GB of storage and it doesn't fit on one EONstor DAS even though they have a capacity of 9 TB each. It's not a perfect solution, I know, but I have to use the hardware we have available. If we loose the backup because of a failure then it's not the end of the world but surely an annoying thing.

One last thing... how do I backup the LVM config

EDIT: What would be really awesome would be if I could move the HBA to the dedicated server, install CentOS and then restore the LVM config and access the backups that are stored on the two EONstor DAS. What might be a problem is that the disk configuration of the dedicated server is very different from the Linux VM so the FC storage would probably be name differently (and not sdb and sdc).

[tsightler]

Backups are actually made automatically and stored in /etc/lvm/backups (current backup) and /etc/lvm/archive (older revisions). You can simply backup this directory to any alternative location. You can also perform manual backups of LVM at any time using the "vgcfgbackup" command and restore those backups using "vgcfgrestore". You probably should read the following sections in the LVM users guide:

http://www.centos.org/docs/5/html/Clust ... ackup.html

http://www.centos.org/docs/5/html/Clust ... ackup.html

Even without a backup it's generally not too difficult to recover the LVM config, but having the backups usually makes the task trivial, unless of course you're trying to recover from advanced failures like bad disk blocks, etc. which is always tricky no matter the volume management involved.

[rfn]

Thank you very much for all the information. I'm actually on vacation right now but will start building the physical machine when I'm back and see if I can move the LVM config to that so that I don't have to start the backups from scratch.

BTW, will the Linux target solution give me better performance than fx. an iSCSI solution by using OpenFiler og FreeNAS (on the same physical machine) and using the MS iSCSI initiator directly on the Veeam VM?

[tsightler]

Usually the linux target will provide better performance since all of the I/O is local, and the TCP stream is simple, however, in cases where the Linux system is undersized this might not always be the case. I have performed extensive testing with using the various methods and found that a RHEL/CentOS 6.x system with a well formatted ext4 filesystem is pretty hard to beat as a target. That being said, I've also done a lot of work with Nexenta, which uses the OpenSolaris kernel and ZFS filesystem as the core and performance is quite good as both iSCSI and CIFS.

[rfn]

Sounds good since I'm using CentOS 6.2 but not ext4 though. Apparently it's not possible to create filesystems larger than 16 TB with ext4 without doing some tinkering so I had to choose XFS. I searched the web for comparisons and it seems like XFS is just fine for our use.

iSCSI is interesting because I would then be able to format the filesystem as NTFS and have no problem using it on a random system, but I'm not sure it that's easy when I group together two DAS.

The first full backup of the large VM is running now and is transferring 100 MB/s (real MB/s, not Veeam artificial) to the Linux VM so performance is very good. Veeam identifies the target to be a 10% bottleneck where the source is more than 90% which is actually quite surprising since the Veeam VM (8 vCPU) is running on fast servers with 8 Gb/s FC access to our HDS HUS 110 SAN, so there should be plenty of bandwith. The Linux VM is running on the same host.