Proxmox VE Network Bonding

Proxmox VE · Linux Networking

Proxmox VE Network Bonding: All 7 Modes Explained

Learn how the seven Proxmox VE bonding modes work, where to use them, and how to configure and verify network redundancy.

What is Linux network bonding?

Linux Network Bonding combines multiple physical network interface cards (NICs) into a single logical interface such as bond0. Depending on its mode, a bond can provide physical-link failover or distribute traffic across several connections. Two 1 Gbps links do not automatically deliver 2 Gbps for a single TCP flow.

A common Proxmox topology is eno1 + eno2 → bond0 → vmbr0 → Virtual Machines. In this bridged configuration, the host’s management IP typically belongs to vmbr0.

Comparison of all seven Linux bonding modes

Mode	Name	Distribution	Link redundancy	Switch requirements
0	`balance-rr`	Sequential outgoing packets	Yes	Usually compatible static aggregation
1	`active-backup`	One active, others standby	Yes	No special aggregation
2	`balance-xor`	Hash-based	Yes	Usually static aggregation
3	`broadcast`	Copies all outgoing packets	Yes	Topology-specific
4	`802.3ad`	LACP + hash	Yes	LACP required
5	`balance-tlb`	Adaptive transmit	Yes	No special aggregation
6	`balance-alb`	Adaptive transmit + IPv4 receive	Yes	No special aggregation

Mode 0 – balance-rr (Round Robin)

How it works: Outgoing packets are sent sequentially through member interfaces: packet 1 over eno1, packet 2 over eno2, packet 3 over eno1 again. This is packet-level distribution.

Advantages: May use several physical paths and provide link redundancy and additional aggregate transmission capacity in compatible setups.

Disadvantages: Packets can arrive out of order, negatively affecting TCP performance. Compatible static link aggregation is usually required at the switch. Per-flow throughput gains are not guaranteed.

Best use: Specialized network configurations tested for ordering and peer-device compatibility. Not the default choice for general production Proxmox networking.

Mode 1 – active-backup

How it works: One NIC carries normal traffic; other NICs remain in standby. A standby takes over when the monitored active link fails. bond-primary selects a preferred interface.

Advantages: Simple, does not require LACP, and can connect to independent switches for redundancy when properly designed. Well suited for Proxmox management.

Disadvantages: Only one interface carries normal traffic at any time. Two 1 Gbps links do not provide 2 Gbps aggregate throughput in this mode. Basic physical link monitoring (miimon) cannot detect every upstream failure.

Best use: Management, small-to-medium virtualization hosts, and deployments where availability is more important than aggregate bandwidth.

Mode 2 – balance-xor

How it works: The transmit interface is selected using a hash. The xmit_hash_policy setting can consider L2, L3 and, for supported traffic, L4 information. Keeping each flow on its chosen link helps preserve packet order.

Advantages: Distributes different flows, supports redundancy, and is more consistent per flow than packet-level round robin.

Disadvantages: Usually requires compatible static port aggregation on the switch. Single-flow speed is generally capped at one member link, and hashes can distribute traffic unevenly.

Best use: Static port-channel environments where LACP is not deployed.

Mode 3 – broadcast

How it works: Every outgoing packet is transmitted on every active member interface. The goal is path duplication rather than load distribution.

Advantages: May serve specialized redundant-transmission designs.

Disadvantages: Uses substantially more network capacity, can create duplicate packets, and does not increase normal VM application throughput. Requires careful topology planning.

Best use: Unusual systems explicitly requiring redundant packet transmission.

Mode 4 – IEEE 802.3ad (LACP)

How it works: LACP manages several Ethernet connections as a logical aggregation group. Outgoing flows are assigned to member links according to a hashing policy. Two 10 Gbps interfaces may supply roughly 20 Gbps of aggregate link capacity when enough parallel flows exist. A single TCP connection typically remains limited to one member link.

Advantages: Enterprise interoperability, dynamic aggregation coordination, link failure handling, and improved aggregate throughput for multiple flows.

Disadvantages: The switch must be configured with a matching LACP group. A bond spanning separate physical switches requires MLAG, switch stacking, or equivalent multi-chassis aggregation. VLAN and hash policies require attention.

Best use: High-density Proxmox hosts, parallel backup/storage transfers and live migration networks. layer3+4 hashing can distribute some flows more efficiently but its packet-ordering properties and 802.3ad interoperability should be validated.

Mode 5 – balance-tlb (Adaptive Transmit Load Balancing)

How it works: Outgoing traffic can be balanced according to interface load and link characteristics. Incoming traffic is generally handled on a selected receiving link. No special switch aggregation is required.

Advantages: Transmit balancing and physical-link resilience without switch-side LACP.

Disadvantages: Receive capacity is normally tied to a single NIC. Bridged VM networks and complex MAC topologies require compatibility testing.

Best use: Validated Linux server scenarios with transmit-heavy workloads and limited switch configuration options.

Mode 6 – balance-alb (Adaptive Load Balancing)

How it works: Adds IPv4 receive load balancing to the transmit behavior of balance-tlb, using ARP-based mechanisms that can direct different IPv4 peers to different physical MAC addresses.

Advantages: Can distribute outgoing traffic and suitable incoming IPv4 traffic without LACP; offers link redundancy.

Disadvantages: IPv6 does not benefit from the same ARP-based receiving mechanism. Port security, ARP inspection, Linux bridges, and multiple guest MAC addresses can create compatibility issues.

Best use: Specialized and thoroughly tested physical Linux host designs, rather than ordinary multi-VM Proxmox networks.

Creating a bond in the Proxmox web interface

Open Node → System → Network → Create → Linux Bond. Set Name to bond0, add eno1 eno2 in Slaves, and choose Mode. Enable Autostart if needed. Assign an IP directly to the bond only when the design requires it; when using a Linux Bridge for virtual machines, the host IP usually belongs to the bridge.

Hash policy influences outgoing distribution for supported modes, while bond-primary is especially relevant for active-backup. Always have independent console or iDRAC/iLO/IPMI access before applying remote network changes.

Example: active-backup with a Linux Bridge

auto eno1
iface eno1 inet manual

auto eno2
iface eno2 inet manual

auto bond0
iface bond0 inet manual
    bond-slaves eno1 eno2
    bond-mode active-backup
    bond-miimon 100
    bond-primary eno1

auto vmbr0
iface vmbr0 inet static
    address 192.168.10.10/24
    gateway 192.168.10.1
    bridge-ports bond0
    bridge-stp off
    bridge-fd 0

Example: 802.3ad LACP with a Linux Bridge

auto eno1
iface eno1 inet manual

auto eno2
iface eno2 inet manual

auto bond0
iface bond0 inet manual
    bond-slaves eno1 eno2
    bond-mode 802.3ad
    bond-miimon 100
    bond-xmit-hash-policy layer3+4
    bond-lacp-rate fast

auto vmbr0
iface vmbr0 inet static
    address 192.168.10.10/24
    gateway 192.168.10.1
    bridge-ports bond0
    bridge-stp off
    bridge-fd 0

Note: Addresses and interface names are examples. For LACP, peer switch interfaces must be members of a matching aggregation group.

Verifying bonding status

cat /proc/net/bonding/bond0
ip -br link
ip -br address
ethtool eno1
ethtool eno2

The bonding status file shows the active mode, members, link state, and mode-specific details. For LACP, verify the LAG status on the switch as well.

Proxmox Corosync and bonding

Corosync needs a stable, low-latency network. Proxmox guidance favors redundant and independent physical paths for cluster communication rather than assuming that one bond solves all network failures. Several load-balancing modes can behave poorly under certain failure conditions. Isolate Corosync from high-volume VM/storage workloads and measure failover behavior before production rollout.

Frequently asked questions

Will two 1 Gbps interfaces give a single 2 Gbps TCP connection?

Generally no. LACP can provide close to 2 Gbps of aggregate capacity across multiple eligible flows, but one TCP flow normally uses only one member link.

Is a managed switch mandatory?

No. active-backup requires no special aggregation support. 802.3ad needs a compatible LACP-capable peer.

Are Linux Bonding and Linux Bridge the same?

No. Bonding groups physical network links into a logical interface; a bridge connects host and virtual machine interfaces at Layer 2.

Does bonding replace Proxmox HA?

No. Bonding addresses physical network connectivity. Proxmox HA handles cluster-level recovery and operation of virtual machines and containers.

Conclusion

For most Proxmox installations, active-backup is a practical first choice for straightforward failover, while 802.3ad (LACP) is a leading choice for higher aggregate capacity across multiple flows when the network equipment supports it. Choose other modes only after evaluating the workload, topology, and failover tests.