Configure Network Bonding for Redundancy on CentOS

Network bonding joins two or more physical network interfaces into one logical link. On CentOS, this can keep a server reachable when a cable, network adapter, or switch port fails. Bonding is useful for web servers, database hosts, virtualisation nodes, and storage systems where a brief network interruption is costly.

Australian administrators commonly use redundant links in Sydney and Melbourne colocation facilities, as well as smaller server rooms in Brisbane and Perth. The design should match the switch configuration, the CentOS release, and the organisation’s security requirements, including controls influenced by the Australian Privacy Act 1988 and the Essential Eight.

Choose the right bonding mode

The active-backup mode is usually the safest starting point. One interface carries traffic while the second waits as a standby. It requires little switch configuration and provides failover if the active cable, NIC, or switch port stops working.

Link aggregation modes such as 802.3ad, also called LACP, can use multiple links at once. They require compatible configuration on the switch, and the switch must normally place both ports in the same bonded channel. A mismatch can cause packet loss or unstable connectivity rather than improving resilience.

Check interfaces and network details

List the available interfaces before changing configuration:

ip link show
nmcli device status
lspci | grep -i ethernet

Record the interface names, current IP address, default gateway, DNS servers, and switch port assignments. Modern CentOS systems may use names such as ens160 or enp1s0; do not assume they are eth0 and eth1.

Keep a console connection available during the change. If the server is remote, arrange out-of-band access through a management controller or a provider console. This is particularly important in Australian colocation environments, where a failed remote network change may require a technician to access the rack.

For broader command-line reference material, the Linux Unix guides provide related administration topics that complement interface and service troubleshooting.

Create a bond with NetworkManager

NetworkManager is the preferred method on current CentOS installations. The following example creates an active-backup bond using enp1s0 and enp2s0:

sudo nmcli con add type bond ifname bond0 con-name bond0 \
  bond.options "mode=active-backup,miimon=100"

sudo nmcli con add type ethernet ifname enp1s0 \
  con-name bond0-slave1 master bond0

sudo nmcli con add type ethernet ifname enp2s0 \
  con-name bond0-slave2 master bond0

Assign the server’s address to the bond, not to either physical interface. Replace the example values with the correct address for the host:

sudo nmcli con mod bond0 \
  ipv4.method manual \
  ipv4.addresses 192.0.2.20/24 \
  ipv4.gateway 192.0.2.1 \
  ipv4.dns "192.0.2.53 1.1.1.1" \
  ipv6.method disabled

sudo nmcli con up bond0

The miimon=100 setting checks link status every 100 milliseconds. It detects a physical link failure quickly, although it may not detect every upstream switching problem. Keep the member interfaces free of separate IP addresses and avoid defining competing default gateways.

Configure LACP when required

For active-active bandwidth and switch-managed aggregation, create the bond with LACP settings:

sudo nmcli con add type bond ifname bond0 con-name bond0 \
  bond.options "mode=802.3ad,miimon=100,lacp_rate=fast,xmit_hash_policy=layer3+4"

The switch ports must be configured as one LACP group, with matching VLAN membership, native VLAN rules, speed, and duplex settings. A server connected to two unrelated switches cannot use ordinary LACP unless the switching platform supports a multi-chassis aggregation feature.

If the server hosts a MariaDB workload, network redundancy should be tested alongside database recovery procedures; the MariaDB on Fedora guide offers useful installation context even when the production host uses CentOS.

Test failover and inspect faults

Confirm the bond state and active member:

cat /proc/net/bonding/bond0
nmcli connection show
ip addr show bond0
ip route

Test from another host with continuous ping, then disconnect the active cable or disable its switch port. Traffic should continue through the backup interface. Restore the link and verify whether the bond returns to the preferred member.

Useful diagnostics include:

journalctl -u NetworkManager -b
nmcli device show bond0
ethtool enp1s0
ip -s link show bond0

A bond does not protect against every failure. It cannot compensate for a failed gateway, an incorrect VLAN, a dead switch stack, or a firewall rule blocking the traffic. Australian businesses aligning systems with the Essential Eight should also monitor link changes, retain relevant logs, and document who can alter switch and host networking.

Compare practical bonding options

The right mode depends on switch support, the failure scenario, and whether extra throughput matters. A simple failover design is often easier to operate than a more complex aggregation arrangement.

When infrastructure documentation covers both Linux services and a content platform, a reference about a custom WordPress block can sit alongside the server runbook, provided access credentials and network diagrams remain protected.

Mode	Traffic use	Switch configuration	Best fit
active-backup	One link at a time	Usually none	Reliable failover with simple operations
balance-xor	Distributed by hash	Often required	Controlled static aggregation
802.3ad	LACP-based distribution	Required	Redundant links with compatible switches
balance-alb	Adaptive transmit and receive balancing	Usually none	Selected environments without LACP

The main principle is simple: place the IP configuration on bond0, make the switch and host agree on the bonding mode, and test a real cable or port failure before relying on redundancy.