Bobay Stack

Networking & proxies / 2026-04-19

vmbr bridges and SDN: which one, and when

The default Linux bridge covers nine out of ten Proxmox networking needs. SDN exists for multi-tenancy and isolation — know the boundary first.

3 MIN READ BY BOBAY.TOP Updated 2026-04-25 ↖ All notes
vmbr bridges and SDN: which one, and when

vmbr0 is the most common name in a Proxmox network config. It is just a Linux bridge putting the physical NIC and the virtual NICs into one layer-2 domain. Simple in theory — and the source of most "the VM has no network" tickets.

Three things to check on a bridge

cat /etc/network/interfaces        # is the bridge bound to the right NIC?
ip link show master vmbr0          # which ports are in the bridge?
qm config 101 | grep net0          # is the guest NIC on that bridge?

A classic failure is the physical NIC being managed by both the bridge and NetworkManager: the bridge comes up, the physical port does not pass traffic. Exclude it:

nmcli device set enp3s0 managed no

Isolated subnet: SDN or a hand-made bridge

When containers need to reach each other but not the main LAN:

ApproachCostFits
Extra vmbr1 with no physical portLow, edit interfacesSingle host, temporary isolation
SDN (Simple zone + VNet)Medium, enable SDN and reloadSeveral nodes, central management

For single-host isolation I add a bridge with no uplink, five minutes of work:

auto vmbr1
iface vmbr1 inet static
    address 10.20.0.1/24
    bridge-ports none
    bridge-stp off

Point the container NICs at vmbr1. It has no way out unless you add routing or NAT — which is exactly the point of isolating it.

When SDN earns its complexity

Three or more nodes sharing subnets, per-tenant VLANs, or a desire to stop editing /etc/network/interfaces and restarting networking. The cost is more concepts (zone, vnet, subnet, ipam) and one more layer to debug. On a single host, SDN is mostly self-inflicted trouble.

A debugging order that takes two minutes

  1. ip a — are the bridge and the physical port UP?
  2. qm config / pct config — which bridge is the guest NIC on?
  3. Inside the guest, ip a — did it get an address, DHCP or static?
  4. On the host, tcpdump -i vmbr0 arp — is ARP leaving?

Step four is the decisive one. No ARP leaving means the problem is at the network layer; ARP leaving with no reply means the guest config is wrong.

What a bridge actually is

vmbr0 is a layer-2 switch: the physical NIC is the uplink, guest NICs are ports. From that, three rules follow — the bridge itself needs no IP unless you manage through it, that IP is a management address rather than a forwarding one, and one physical NIC belongs to one bridge unless you bond.

Minimal working config

auto vmbr0
iface vmbr0 inet static
    address 192.168.1.10/24
    gateway 192.168.1.1
    bridge-ports enp3s0
    bridge-stp off
    bridge-fd 0

Apply with ifreload -a, not systemctl restart networking — the latter drops every interface at once and takes your SSH session with it. Layer two first, layer three second: if brctl show does not list the guest's veth, the problem is the guest's NIC config, not the network.

PVEnetworkingbridgeSDN

References

PVE SDN docs ↗