Article ID: 117979, created on Oct 15, 2013, last review on Oct 15, 2014

  • Applies to:
  • Virtuozzo
  • Virtuozzo containers for Linux
  • Virtuozzo hypervisor


This article describes which bonding modes are supported by Parallels virtualization products for virtual environments.

The information is provided for configurations with a single network segment, when interfaces of a server are connected to the same network switch, or when interfaces are connected to different switches which are connected with each other, so that there is a single network segment.


Parallels virtualization products allow to attach virtual environments in the bridged mode to Ethernet networks. For hardware servers with several Ethernet interfaces, it is possible to aggregate few interfaces to a single bonding interface to increase bandwidth and/or provide fault-tolerance.

Technically, a virtual network is represented by a bridge (the interface brX on the hardware node) with one physical "master" interface. It is possible to attach an aggregated interface to the bridge, for example - a bonding interface. Node administrators should pay special attention if there is a bonding interface specified as a physical interface for a virtual network.

Bonding modes supported on the hardware node

The following bonding modes are supported by the kernel module:

  • Mode 0 (balance-rr - round-robin policy) transmits network packets in sequential order from the first available network interface (NIC) slave through the last. Mode 0 provides load balancing and fault tolerance.
  • Mode 1 (active-backup - active-backup policy) sets all interfaces to the backup state while one remains active. Upon failure on the active interface, a backup interface replaces it as the only active interface in the bond. The MAC address of the bond in mode 1 is visible on only one port (the network adapter), to prevent confusion for the switch. Mode 1 provides fault tolerance.
  • Mode 2 (balance-xor - XOR policy) selects an interface to transmit packets to based on the result of an XOR operation on the source and destination MAC addresses multiplied by the modulo slave count. This calculation ensures that the same interface is selected for each destination MAC address used. Mode 2 provides fault tolerance and load balancing.
  • Mode 3 (broadcast - broadcast policy) transmits network packets on all slave network interfaces. Mode 3 provides fault tolerance.
  • Mode 4 (802.3ad - IEEE 802.3ad policy) creates aggregation groups for which included interfaces share the speed and duplex settings. Mode 4 uses all interfaces in the active aggregation group in accordance with the IEEE 802.3ad specification.
  • Mode 5 (balance-tlb - adaptive transmit load balancing policy) ensures the outgoing traffic distribution is according to the load on each interface and that the current interface receives all incoming traffic. If the interface assigned to receive traffic fails, another interface is assigned the receiving role instead.
  • Mode 6 (balance-alb - adaptive receive and transmit load balancing policy) includes balance-tlb plus receive load balancing (rlb) for IPV4 traffic, and does not require any special network switch support. The receive load balancing is achieved by ARP negotiation. The bonding driver intercepts the ARP replies sent by the local system on their way out and overwrites the source hardware address with the unique hardware address of one of the NIC slaves in the single logical bonded interface such that different network-peers use different MAC addresses for their network packet traffic.

Bonding modes incompatible with bridging

Interfaces operating in these modes cannot be used as a physical interface in virtual networks by design of these modes:

  • Mode 3 (broadcast - broadcast policy) is inefficient, since it sends the same traffic through all slave interfaces;
  • Mode 6 (balance-alb - adaptive receive and transmit load balancing policy) replaces source MAC addresses in all outgoing packets and this blocks the possibility to use DHCP in virtual environments.

Bonding modes available for bridging without special configuration

The only mode that does not need any special configuration on the hardware node or a network switch is:

  • Mode 1 (active-backup - active-backup policy).

This mode uses a single interface to send and receive packets, ignoring the traffic received through inactive slave interfaces. This is important in the bridged mode.

Technical explanation of the situation

The main point is to avoid receiving ARP announcements back in situations when such announcement is sent from a virtual environment. The packet flow for the announcement about an IP address assignment to a virtual environment is:

  1. an ARP packet is sent to the ethX interface in the virtual environment;
  2. this packet is handled in the interface vethN.M (for a container) or vmeN.M (for a virtual machine) on the hardware node side;
  3. since the environment is configured in the bridged mode, the interfaces vethN.M and vmeN.M are connected with the bridge brN on the hardware node, to a certain bridge's port and therefore the bridge auto-learns that the MAC address of the virtual environment is accessible through this port;
  4. the ARP announcement (being a broadcast message) is sent by the bridge (acting as a unmanaged switch) to all other ports of this bridge - to the bonding interface bondN as well, among all attached interfaces;
  5. the bonding module uses some slave interface to send the packet through in accordance with the current policy and the packet reaches the network switch which the node is connected with;
  6. the network switch does almost the same as the bridge - it sends the packet to all other active ports;
  7. without proper configuration, the switch can send the packet back to the hardware node through other interfaces included into bondN;
  8. if the hardware node accepts packets from all slaves, it receives back the recently sent ARP announcement and rewrites the location of the MAC address forgetting the proper port (learnt on the step 3 above) and marks this MAC address as available in the external network.

The only mode which works without any additional configuration is active-backup, because packets from inactive slave interfaces are not processed.

Bonding modes supported with a proper network switch configuration

The following modes require that the interfaces are aggregated on the network switch side to avoid looping ARP announcements back to the node:

  • Mode 0 (balance-rr - round-robin policy) [1];
  • Mode 2 (balance-xor - XOR policy);
  • Mode 4 (802.3ad - IEEE 802.3ad policy);
  • Mode 5 (balance-tlb - adaptive transmit load balancing policy).

The configuration can be either manual (for a managed switch) or automatic (if the switch supports LACP).

[1] Not all switches support this mode properly.

Additional configuration

It might be necessary to disable offload features on the network interfaces included in the bonding interface. The cause of this is bulk processing of packets by the network driver and packets may be delivered to a wrong environment.

The following command can be put to a script which is executed on the system start:

~# for iface in $(ip link | awk -F ': ' '$3~/\<SLAVE\>/{print $2}'); do ethtool -K $iface tx off rx off tso off gro off gso off; done

Search Words

networkins is not working for VMS on PCS node that has bonded network




bonding options

dropped packets on veth interface


No network

bonding mode

VERY Poor network performance




bridged windows



a26b38f94253cdfbf1028d72cf3a498b 2897d76d56d2010f4e3a28f864d69223 d02f9caf3e11b191a38179103495106f e8e50b42231236b82df27684e7ec0beb 0dd5b9380c7d4884d77587f3eb0fa8ef

Email subscription for changes to this article
Save as PDF