Master Linux Multi‑NIC Bonding: Modes, Configuration Steps & Best Practices

Linux NIC Bonding Overview

Linux kernel provides a bonding driver that can aggregate up to seven physical network interfaces into a single logical interface (e.g., bond0). The most frequently used modes are 0 (balance‑rr), 1 (active‑backup) and 6 (balance‑alb).

Bonding Modes

mode=0 – balance‑rr : round‑robin transmission across slaves; preserves packet order per slave but can cause out‑of‑order delivery for a single TCP flow. Requires switch support for link aggregation (static or LACP).

mode=1 – active‑backup : only one slave is active; others are standby. Provides fault‑tolerance without load‑balancing. The bond presents a single MAC address taken from the active slave, avoiding switch MAC conflicts.

mode=2 – balance‑xor : selects a slave based on an XOR hash of source/destination MAC (or IP) addresses; offers load‑balancing and fault‑tolerance. Hash algorithm can be changed with xmit_hash_policy.

mode=3 – broadcast : every packet is sent on all slaves; maximizes redundancy but does not balance load.

mode=4 – 802.3ad (LACP) : implements IEEE 802.3ad link aggregation; requires switch support and matching LACP configuration on both ends. Each slave must report speed/duplex via ethtool.

mode=5 – balance‑tlb : adaptive transmit load‑balancing; distributes outgoing traffic according to each slave’s current load. No special switch configuration needed.

mode=6 – balance‑alb : combines TLB with receive load‑balancing (RLB) using ARP negotiation; each slave uses a unique MAC address, so no switch configuration is required.

Key Parameters

mode

: bonding mode (0‑6). miimon: link‑monitoring interval in milliseconds (e.g., 200 ms). max_bonds: maximum number of bond interfaces that can be created. xmit_hash_policy: hash algorithm for mode 2 (default XOR) or mode 4. updelay and downdelay: delay before bringing a slave up or down; important for ARP‑based modes.

Configuration Example for a Single Bond (bond0)

Create the bond interface definition /etc/sysconfig/network-scripts/ifcfg-bond0 Typical content:

DEVICE=bond0
BOOTPROTO=none
ONBOOT=yes
IPADDR=192.168.0.100
NETMASK=255.255.255.0
NETWORK=192.168.0.0
BROADCAST=192.168.0.255

Configure each slave interface Example for eth0 :

/etc/sysconfig/network-scripts/ifcfg-eth0

DEVICE=eth0
BOOTPROTO=none
MASTER=bond0
SLAVE=yes

Repeat the same file for eth1 (change DEVICE ).

Set bonding module options Create /etc/modprobe.d/bonding.conf with the desired mode and monitoring interval:

alias bond0 bonding
options bonding mode=0 miimon=200

Load the module (or reboot): modprobe bonding Restart networking and verify /etc/init.d/network restart Check the bond status:

cat /proc/net/bonding/bond0

Multiple Bonds

If more than one bond is required, the bonding.conf file can apply a common set of options to all bonds or specify per‑bond options.

Uniform mode for all bonds

alias bond0 bonding
alias bond1 bonding
options bonding max_bonds=2 miimon=200 mode=1

Different modes per bond

alias bond0 bonding
options bond0 miimon=100 mode=1
install bond1 /sbin/modprobe bonding -o bond1 miimon=200 mode=0

Verification Output Example

Typical /proc/net/bonding/bond0 excerpt:

Ethernet Channel Bonding Driver: v3.5.0 (November 4, 2008)
Bonding Mode: fault-tolerance (active-backup)
Primary Slave: None
Currently Active Slave: eth0
MII Status: up
MII Polling Interval (ms): 200
...

The bond and all slaves share the same MAC address in active‑backup mode, preventing switch MAC‑flapping.

Practical Caveats

Mode 0 requires the connected switch ports to be configured as an aggregated link (static or LACP) because all slaves use the same MAC address.

Mode 6 (balance‑alb) rewrites ARP replies so each slave presents a distinct MAC address; no switch configuration is needed.

When adding or removing slaves, set updelay to a value larger than the switch’s forwarding delay to avoid temporary loss of ARP entries.

Use ethtool to verify that each NIC reports speed and duplex; this information is required for LACP (mode 4) and for accurate load‑balancing in modes 5 and 6.