CPU/RAM recommendations for 50TB all-in-one VBR server

[tore]

I am looking for advice on how to spec a new Veeam server in regards to CPU and RAM.

We are backing up about 80 VMs (local vSphere) and 4 physical server (WAN).
We have 4 ESXi hosts and a 10Gbe iSCSI SAN. The backup server will transport data directly from the SAN.

Total amount of data is about 20TB, or 10TB actually stored data (compressed and deduped).

I have a quote for a server with:
CPU: 1 x Xeon Silver 4112 - 4 core 2.6GHz
RAM: 2 x 16GB
OS disk: 2 x 240GB SSD
Repository: 10 x 10TB NL-SAS drives which will be set up with RAID10

The server will be a all-in-one Veeam Enterpise backup server, including tape server (SAS connected LTO8 autoloader).
I am thinking Win 2016 and ReFS for the repository.

Will this CPU and RAM suffice?

[PTide]

Hi,

If the server is supposed to be the only server in the backup infrastructure, therefore combining proxy and repository roles, then please refer to this article and sum up requirements for separate components.
Also there is the best practices guide which is worth looking at.

Thanks

[tore]

Thank you!

Yes, this will be the only server combining proxy and repository. But we will possibly extend with combinded proxy/repository servers at branch offices for ESXi's in these branch office. We might want to copy backup to the central site where this new server will be located.
All these 80VMs, 20TB, is in one job with per-VM backup files.

RAM is easy and inexpencive to upgrade, and also requirements can be easy to sum up.
Processor needs is what I am most unsure about.
Will 4 cores be enough? Should I up the core count with lower clock speed at approximately the same cost?

[jmmarton]

The clock speed in modern processors isn't all that relevant these days as they're all fast. What's most important is core count. As long as the core count lines up with what's in our documentation (and especially the best practices guide) for all Veeam components on the box, the clock speed of the CPUs in any new server being purchased today should suffice.

Joe

[nitramd]

A Veeam proxy uses 1 physical CPU core for each VM disk. For sake of illustration, if each of your VMs has one disk then you would need 80 cores. However, VMs usually have more than one disk hence the need for more cores.

In reality, we're all constrained by budget, i.e. we purchase the most performance for what we can afford, ergo, lower physical core count per CPU. I would suggest using virtual proxies, they work fairly well; if this is possible. Think of a Veeam proxy as a data mover. Additionally, I would suggest obtaining the total number of disks your VMs have.

Also, don't forget about the physical core count on your repository - more cores, more better; think ingest.

Also also, if you're going to back up physical servers take a good look at Veeam Agent for Windows (VAW) or Veeam Agent for Linux (VAL). This is the only way to go for backing up physicals.

Note that this post is a short, high level overview (sans a lot of detail) but it should point you in the right direction.

Lastly, if you know of a chip manufacturer that has an 80 core, general purpose CPU let me know ASAP, I'll budget for some.

Good luck.

[pkelly_sts]

One method we use is, as has been mentioned, a virtual proxy on each host, as well as the main repository server with direct-attach storage.

Each of these vProxies in our environment has been configured with 6 cores, so effectively 6 simultaneous backup "objects".

My reasoning is that, in our case, all backups are done at night and we're not really a 24/7 shop so I'm quite happy for the vProxies to use up the otherwise little-used CPU capacity at night time.

As a result of this config our 60+ VMs are generally backed up in <30 mins each run.

This may or may not fit with your business needs but thought I'd put it out there...

[RGijsen]

Tore,
how did this setup work out in terms of performance? With 10 spinning disks and ReFS, I assume with FastClone enabled, are you able to satisfy your tapedrive for example? We are running 16*1TB (2.5'') NL-SAS in a dedicated SAN, although in a RAID6 (I know, it kills write performance) but now after half a year of ReFS magic, restores run extremely slow because of al that random IO block-cloning introduced.

So how did it work out for you?