Roughly how many VMs before physical proxies?

[mikeely]

I know best practice is to put the backup proxies on physical hosts and understand the reason why. What I'm wondering is at about what point in terms of VMs backed up do I want to make that change? Also, if there's a sizing document to help spec out the physical server needed that'd be useful.

[Stokkolm]

https://www.veeam.com/pdf/guide/veeam_b ... ere_en.pdf

See Systems Requirements Section where it talks about number of CPU cores & memory per backup job.

[JoshuaPostSAMC]

The primary reason for physical proxies is to get Direct SAN/NFS and remove the backup traffic from the VM traffic, right?

We have Cisco UCS servers for VMware with 40 GB networks between the blades, and multiple 10 GB going out the fabric interconnects, so I don't worry about the network impact of sending traffic over ethernet out to my CIFS repository, but if you had limited bandwidth, I could see the case for it.

I have 530 VMs I'm backing up, 7 virtual proxies with 8 vCPU and 16 GB RAM each, and my Backup server with 8 vCPU and 32 GB RAM and a CIFS repository. I usually push around 2 Gbps traffic out to the repository and after upgrading to 9.5 I'm seeing even more. Right now I'm doing some Active Full backups to enable Per-VM backup files, and I'm getting 8 Gbps on the backup server.

[mikeely]

https://www.veeam.com/pdf/guide/veeam_b ... ere_en.pdf

See Systems Requirements Section where it talks about number of CPU cores & memory per backup job.

Thanks for that. I was also hoping to see what people are experiencing out in the real world and was happy to see that below.

Sure would be nice if the backup proxies could run on something not requiring the purchase of a Windows license...

[mikeely]

I have 530 VMs I'm backing up, 7 virtual proxies with 8 vCPU and 16 GB RAM each, and my Backup server with 8 vCPU and 32 GB RAM and a CIFS repository. I usually push around 2 Gbps traffic out to the repository and after upgrading to 9.5 I'm seeing even more. Right now I'm doing some Active Full backups to enable Per-VM backup files, and I'm getting 8 Gbps on the backup server.

Appreciate the reply. We're 10 GB throughout so I'm not especially concerned about network saturation. I am a bit concerned about slagging the CPU on the hosts though. How's CPU and memory load for you?

[Stokkolm]

It's very CPU core intensive. I only have one VMware cluster and I wouldn't run it there. Having said that, if you're cluster CPU utilization is pretty low you might be able to get away with it.

[JoshuaPostSAMC]

Backup server gets hit pretty hard on CPU and RAM, although I made the CPU better by doing the proper exclusions from Anti-virus so System Center Endpoint stopped scanning all of the logs files all of the time.

Proxies are a little harder to tell as I never look at them during a job. Looking at performance stats in vCenter for a week view (which skews things) shows 10% or lower pre-9.5, and up to around 20% after, but my backup speed has increased significantly. I have all jobs set to Optimal Compression and LAN storage optimization which helped with my jobs. I"ll eventually try going back to High or Extreme compression.

We run backups at night when I have plenty of compute resources, so I'm not too worried about contention in my environment.

I'm not sure how you are licensing Windows, but if you have high consolidation rates, you probably should be doing a Datacenter license for each processor in your virtual environment so you can have as many VMs in that host as they can handle, so you won't be paying extra for a Windows proxy if they are virtual. If you go physical, then that is something to take into consideration.

[mikeely]

I'm not sure how you are licensing Windows, but if you have high consolidation rates, you probably should be doing a Datacenter license for each processor in your virtual environment so you can have as many VMs in that host as they can handle, so you won't be paying extra for a Windows proxy if they are virtual. If you go physical, then that is something to take into consideration.

Given that we're almost exclusively a Linux shop, the delta between where we are and the break-even point for licensing Datacenter is pretty wide. It'd be kind of ironic if we had to fork over that kind of cabbage to Microsoft just to back up our Linux infrastructure...

[JoshuaPostSAMC]

Well, that makes sense then. Still could be a better trade off to license one Vmware host with Datacenter rather than a several physicals with Standard.

[infused]

I run them in my vmware cluster as vms. I have 4gbit between hosts to cifs share. cpu in the cluster is ~5% at night so it's no issue for me. Around 60vms and I run two proxies with 2cpu, 8gb memory.

The proxies get maxed out.

[DeadEyedJacks]

Hi,
With a 10Gb LAN or better, the source storage is probably going to be your bottleneck.
What type of storage and which connectivity protocols are being used?
Are you monitoring and managing the storage latency?

Next the VMware infrastructure becomes a limiting factor.
Each vSphere host cannot safely process more than 7 VMDKs concurrently.
each VMFS datastore will by default be limited to 4 concurrent snapshots of VMDKs.

So the number and arrangement of VMDKs, VMFS datastores, storage and host clusters can become a factor.

Just to say with 600+ VMs and ~1800 VMDKs, its not being the number of proxies, their size or target storage performance limiting our backup throughput.

[Gostev]

Each vSphere host cannot safely process more than 7 VMDKs concurrently.

Note that this limit has been lifted in the most recent release, see What's New in 9.5 for more details!