Backup repository HW requirements

[subglo]

Hi guys,

Please suggest hardware requirements for backup repository for our scenario and/or any tweaking tips.

Using latest VBR 9.5 Update 4.

Currently our physical Windows server 2016 has 2 Intel Xeon CPU's X5675 @3.07Ghz.

In total 12 cores and 24 logical processors (hyper-threading). 32GB RAM.

When we are sending simultaneously 4 VM active full backups (each full 1TB data) trough hotadd proxies to this repository we are getting on average 5.5 GBps (687.5 MB/s) transfer receive speed, but all 4 Veeamagents there are using CPU 100%.

This backup repository has no other Veeam components other than datamover itself.

We would try to achieve maybe even better speeds, but it seems that CPU's can't handle the load. Before we had worse results and we upgrade CPU's which helped, but we have 10 GB uplinks, so there is room for improvement

As per documentation there isn't nothing much for requirements.

CPU: x86 processor (x86-64 recommended).
Memory: 4 GB RAM, plus up to 2 GB RAM (32-bit OS) or up to 4 GB RAM (64-bit OS) for each
concurrent job depending on backup chain’s length and backup files sizes.

[Vitaliy S.]

Hi Aldis,

If you want to improve backup job performance, then you need to take a look at the bottleneck stats first. If the bottleneck is in the source storage, then making changes to the repository will not bring you benefits.

Thanks!

[subglo]

Bottleneck is network in all our latest tests. Not source disks and not destination disks. As i previously stated, then upgrading CPU's on backup repository increased backup and restore speeds by at least 1GBps.

Backup:
avarage - 5.5 Gbps = 687.5 MB/s
peaks - 6 Gbps = 750 MB/s

Restore:
avarage - 3.5 Gbps = 437.5 MB/s
peaks - 4 Gbps = 500 MB/s

[HannesK]

Hello,
as you mention, the CPU of the repository is on 100% load. Which file system are you using? I can remember when I did ReFS tests when it came up, I realized that CPU load was higher on ReFS than with NTFS. I also had a quite old CPU, so maybe that could be the issue?

Best regards,
Hannes

[subglo]

I'm using NTFS.

I'm currently working with support to try solve this. What i'm noticing that for 4 jobs, there are 4 Veeamagent.exe, everyone is using about ~20 CPU in total 80 CPU + 20 CPU for Windows System Interrupts which is bad..

We have 2 identical servers and they act the same. Windows OS and drivers are up do date. As per Google drivers, BIOS,firmware are first ones to look for high System Interrupts usage. As soon as i stop these large backup tests then everything goes to normal.

Well these Intel Xeon CPU's X5675 are from 2011...

[Vitaliy S.]

Bottleneck is network in all our latest tests. Not source disks and not destination disks. As i previously stated, then upgrading CPU's on backup repository increased backup and restore speeds by at least 1GBps.

Upgrading hardware of the components not reported as a bottleneck might increase the performance, but not significantly. If the network connection is reported as the "main guy" here, then you should be looking at upgrading the data link.