Migration from ESXi to PVE: Designing a dedicated 10G LAN

[tedew]

Hello,

We are planning a migration of our virtualization environment from VMware ESXi to Proxmox VE and need advice on designing a dedicated backup network. We want to keep Veeam Backup & Replication as our primary backup solution, using its native integration with Proxmox VE.

Our current ESXi installation:
3 x Production ESXi hosts connected via FC directly to an IBM FlashSystem 5200.
Our backup server is a physical machine (also running ESXi) connected via FC to the same IBM FS5200, allowing Veeam to use magic to transfer large files... This backup host then saves data to a secondary storage array IBM Storwize V5000e via iSCSI.

Our target Proxmox VE setup:
Since the Direct SAN/FC approach is VMware-specific, we are moving to a network-based backup transport over a dedicated LAN.

The 3 production PVE nodes will keep their FC connections to the IBM FS5200 for production storage.

The Veeam backup server will be on PVE instead of ESX, still direct iSCSI connection to the IBM V5000e repository.

To connect the production cluster with the Veeam server, we want to build a completely separate 10G network using 2 ports per node. Production/client traffic will stay on current existing 1G Infra.

Question:
We need high availability (surviving a 10G switch failure) and want to maximize throughput by using both 10G ports simultaneously on each node during the backup windows. We are choosing between:

Option 1: Two standalone 10G switches (No stacking) using a Linux Bond in Balance-ALB (Mode 6) in PVE.

Option 2: Two 10G switches in a physical hardware stack (for example Netgear M430) using a Linux Bond in LACP (Mode 4).

Given that this network will purely handle heavy Veeam backup streams between PVE production nodes and the PVE Veeam repository host, which network design (Balance-ALB vs. Stacked LACP) proves to be more stable, performant, and recommended for PVE?


Thank You,
BR,

 

[crawrj]

I am new to Proxmox so I will let someone else say which one they believe is the best option, but I will say that I am using and have always preferred Stacking with LACP. That aside, why would you put backups on 10G and everything else on 1G? Veeam does incremental so unless you are constantly backing up massive amounts of data 10G, or 60G in your two per host scenario, would be overkill. And even if you are doing that much data then that means the VM's are passing that much data and would warrant 10G more than backups would. Just a thought.

 

[cyruspy]

That's an oddball. I have several doubts on option 1. Go with option 2 out of the two.

With 1GbE, I would separate management to avoid losing access when a workload misbehaves. Also, take into account that VM migrations will probably saturate 1GbE.

In my hyperconverged setups this is my goto:
A: 2 x 25GbE --> management
B: 2 x 25GbE --> workloads + SDN underlay + BGP integrations
C: 2 x 25GbE --> Ceph access
D: 2 x 25GbE --> Ceph replica

Poor man's version: 10GbE. 1GbE is not that usual in the datacenter these days...

Notes:
C/D don't apply in your case.
Backup would use A or B depending on how the solution works (A when connecting to the hypervisor, B when connecting to a "helper VM").