DirectSAN configuration

[Danne84]

Hi,

I'm currently working on a new backup server with DirectSAN access and i would like to know if i need any virtual backup proxies on my hosts when using this transport mode?
We are using revere incremental on our backup jobs and we have mixed VMs with thin and thick disks.

The new machine is physical of course with FibreChannel connection to the SAN.
The repository is local storage.

Also thankful for any suggestions that may improve this installation.

[Pat490]

No you do not need virtual proxy. We have same setup here. 1 physical backup server only with local storage as primary repository!

[Danne84]

I was unsure when i was reading this,

"Direct SAN-based restore is possible only for thick-provisioned VM disks"

what happens when i try to restore VMs with thin disks?

[BartP]

The restored disk will be thick provisioned.
Besides a virtual proxy, you can also restore using NBD mode (if you wish) and restore the VM with thin disks.

[foggy]

what happens when i try to restore VMs with thin disks?

Disk will be restored via network or hotadd (if available) mode.

[Danne84]

Ok so in that case it would be good to have proxys so that they take the load, or am i way off here?
Just trying to get a hang of the restore when using Direct SAN.
What i have to think of, In this case we will backup around 400 VMs and also using Copy-jobs to another site. So when running restore we want it to keep up and not but unnecessary load on the network.

[foggy]

Some of the existing proxies will be used for restore, you can explicitly specify it upon restore.

[Danne84]

Did did some tests and the full backups ran nice and fast, with speeds of over 700MB/s.
The problem came with the increments.
The speed is now down to 50-70MB/s

Can i speed up the incrementals in some way?
I'am using Reverse incremental

Also, did not think about this at first, But is it best practice to have the backup server in the same subnet as the VM?
In this case i'm backing up several VMs that is located in different subnets, So i was thinking if i add an IP for every subnet then veeam backup should be able to work more efficient?
Because now is says "Using guest interaction proxy Veeam01 (Different subnet)"

[mpasaa]

I no longer use direct SAN BUT did have this configured initially and the backups were super fast as expected. Restores are ALWAYS over the network so don't expect restores over fiber or whatever medium the backup was done. As for proxies, I understand proxies are really only used for the READ process when backups start. Remember, no data can be actually written since your physical image job is being written to your Veeam server (or LUN attached to it) ....if that makes sense. Proxies are just to take the load off. I use a bunch of Virtual machines as my proxies in different subnets too. Works like a champ AND if you've got enough virtualization resources you can create proxies with plenty of CPUs to allow for multiple jobs to run.

Direct SAN was OK but I prefer the virtual Veeam server with an attached RDM as it runs quite fast and is more flexible than having to deal with a physical blade which is why I used previously. I also didn't like that the Windows Veeam server was also part of the VM storage groups on the SAN --- it worked and never gave us problems BUT call me paranoid but I was always concerned that one day Windows was going to decide to initialize the data stores or write some disk signature to the VMware datastores which are seen as unknown in Windows.....Veeam was pretty safe though and that's just my paranoia but still...using a Veeam VM is much more flexible for us.

Hope this helps...

[foggy]

Can i speed up the incrementals in some way?
I'am using Reverse incremental

If the bottleneck is target, try switching to forward incremental as the first and easiest step.

Restores are ALWAYS over the network so don't expect restores over fiber or whatever medium the backup was done. As for proxies, I understand proxies are really only used for the READ process when backups start.

That's not true, you can use both direct SAN and hotadd for restores as well.

[mpasaa]

OK...perhaps they've changed some features but that's not what we were ever told when it came to restores and the speed alone clearly showed network speeds as opposed to anything over our fiber network. I only got fiber speeds with backups and I even asked support why don't restores happen over the same medium that was used to backup the virtual machines. Was always told restores always over network mainly because, at the time, Veeam reasoned that because they don't know what type of infrastructure a customer might have the only common option was network mode for restoring. Again, if that's changed I stand corrected but direct SAN is a non-issue for us anyway and is too limiting in our environment. I'd rather have as much virtualization as possible and NOT be tied to physical blades anymore. Much better for us. Thanks for the correction if that has changed. Good to know.

[mpasaa]

This is what we use and while Veeam can restore over certain modes to do so directly to the datastores has specific requirements and certain types of disks must go over network/VA appliance mode (i.e. thin disks which are used frequently) "If the backup proxy is virtualized and resides on the ESX(i) host to which the VM must be restored, Veeam Backup & Replication uses the Virtual appliance transport mode. The Virtual appliance mode utilizes VMware ESX(i) capabilities of HotAdding disks to the VM and thus eliminates the need to transfer the backup data across the network. Veeam Data Movers deployed on the backup repository and backup proxy retrieve VM data from the backup file and put it directly to the necessary datastore via the ESX(i) I/O stack"

[itfnb]

Having a virtual proxy can also be nice to use HotAdd for backups of any VMs that are not stored on your primary SAN. We use this to backup our SAN's management VM that runs on one of our ESXi host's local storage.

[Danne84]

The bootleneck is Source.
I have looked into this and cannot understand why because there is no high load on the SAN. (Fujitsu DX200)
And using Emulex 8Gbit card in the physical server.

But overall the jobs runs verry smooth, Mutch faster than the old setup with virtual VBR server and proxies.

The only drawback is the restore using the network instead of DirectSAN (thin disks)

[foggy]

Do you see the [san] tag in the job session, right after the proxy server name?

[Danne84]

Sorry for late respons,

yes i see the SAN tag, "Using backup proxy VMware Backup Proxy for disk Hard disk 1 [san]"

And also this "Hard disk 1 (60,0 GB) 541,0 MB read at 5 MB/s [CBT][fast clone]"

[Danne84]

I was looking at the backupserver right now and i see that the speed has dropped allot! since i set it up.
Now the jobs are down to 5-10 MB/s when i had around 5-800MB/s when it was setup, and this is only 2 months ago.

Seems to be something wrong here with the performance.

[foggy]

I'd ask support for assistance in investigating the reasons of the low performance.