Advanced Network Configuration and Bonding

Network bonding, also called link aggregation or NIC teaming, is a Linux kernel feature that lets you group several physical network interface controllers (NICs) — the hardware ports on the back of your server — into a single logical interface. To the applications running on the server, it looks like one connection with the combined speed of all the cables. If you have two 1 Gbps NICs bonded, the server sends and receives at up to 2 Gbps, and it keeps working if one cable is unplugged.

Before bonding existed, system administrators had two unattractive options. They could buy expensive, faster network hardware — say, one 10 Gbps NIC — or they could accept the risk that a single cable failure would take the server offline. Bonding solves both problems at the cost of extra cables and switch configuration. The kernel implements bonding through a kernel module called "bonding.ko", loaded with the command "modprobe bonding" or by configuring it in the file "/etc/modprobe.d/bonding.conf". Once loaded, you create a virtual interface — typically named bond0, bond1, and so on — and assign the physical NICs as slaves to that bond interface.

The behaviour of the bond is controlled by its mode parameter. The kernel supports seven modes, numbered 0 through 6. You must memorise at least the four tested by LPIC-2: modes 0, 1, 4, and 6.

Mode 0, balance-rr, stands for round-robin. It transmits packets in sequential order across the slave interfaces. The first packet goes out on the first slave, the second on the second slave, the third on the third slave, and then it cycles back. This provides load balancing and fault tolerance. The big drawback is that packets can arrive out of order on the receiving end, which can confuse some protocols. LPIC-2 loves to test that mode 0 requires switch support — the switch must be configured for link aggregation, or you will have network problems.

Mode 1, active-backup, is the simplest mode for fault tolerance. Only one NIC is active at any time. The other NICs sit idle, waiting for the active one to fail. If the active cable is disconnected or the NIC stops responding, the kernel automatically switches traffic to the next available slave. There is no load balancing. The IP address and Media Access Control (MAC) address stay the same, so there is no disruption to existing connections. This mode works with any standard switch; you do not need a switch that supports link aggregation. LPIC-2 exam questions often present a scenario where maximum reliability with zero performance gain is needed — that is mode 1.

Mode 4, 802.3ad, also called IEEE 802.3ad dynamic link aggregation, is the most complex mode. It uses the Link Aggregation Control Protocol (LACP) to negotiate with the switch to combine ports into a single logical trunk. Both the server and the switch must be configured correctly. LACP packets are exchanged to verify that both sides agree on which ports are part of the bond. Mode 4 provides both load balancing and fault tolerance. The load balancing is based on a hash algorithm — typically a hash of the source and destination MAC addresses or IP addresses — that ensures traffic from the same conversation always uses the same slave, preventing packet reordering. LPIC-2 tests your understanding that mode 4 requires a switch that supports LACP, and that you must configure "lacp_rate" parameters to control how often LACP packets are sent.

Mode 6, balance-alb, stands for adaptive load balancing. It does not require any special switch configuration. It achieves load balancing by dynamically adjusting the MAC addresses of the NICs and by sending Address Resolution Protocol (ARP) replies to redirect incoming traffic to the current transmitting slave. This mode provides both load balancing and fault tolerance. LPIC-2 tests that mode 6 is the only bonding mode that can send and receive load balancing without switch support, but it is also the most CPU-intensive because the kernel software has to handle the MAC address rewriting.

To make a bond interface persistent across reboots on most Linux distributions, you edit network configuration files. On Debian-based systems (like Ubuntu), you edit "/etc/network/interfaces" to define a bond interface with slaves, mode, and other options such as "miimon" (monitoring interval in milliseconds). On Red Hat-based systems (like CentOS or RHEL), you create or edit files in "/etc/sysconfig/network-scripts/" — for example, "ifcfg-bond0" and "ifcfg-eth0" with the "MASTER=bond0" and "SLAVE=yes" directives. The key parameters you need to know for the exam are:

A common beginner mistake is forgetting to load the bonding module before configuring interfaces. If you reboot and the bond interface fails to come up, it is often because the bonding module is not listed in "/etc/modules" or a similar kernel module configuration file. LPIC-2 expects you to know how to load the module with "modprobe bonding" and to ensure it loads automatically.

Bonding is not the same as bridging or routing. Bridging connects two network segments at layer 2 (data link layer) to form a single network. Bridging and bonding are sometimes confused. Bonding combines multiple links into one logical link for redundancy and bandwidth; bridging combines multiple networks into one logical network. The exam may present a scenario where you need to decide which to use. Bonding is purely about the physical link between your server and the switch. It does not change the network topology — it just makes the link faster and more reliable.

Imagine you are the sole IT administrator for a mid-sized e-commerce company. The company runs its online shop on a single Linux server in the office. The server has two NIC ports, each connected to a different network switch. The switches themselves are connected to the company's router and then to the internet. Currently, only one NIC is active. The boss asks: why does the website sometimes become unreachable when a cleaner accidentally bumps the cable? And can we make it faster without buying a 10-gigabit switch?

Your solution is network bonding. Here is exactly what you do:

First, you check the current network configuration. You run "ip link show" and see eth0 and eth1 as the two physical interfaces. Both are 1 Gbps. You have an old switch that does not support LACP, so you cannot use mode 4. You also cannot trust that both cables will always be connected, so you need fault tolerance. But the boss also wants some bandwidth improvement. You decide to use mode 6 (balance-alb) because it works with any switch and gives both load balancing and fault tolerance.

You install the bonding modules. You edit "/etc/modprobe.d/bonding.conf" and add a line to load the bonding module with the options you need: "alias bond0 bonding" and "options bonding miimon=100 mode=6". On a Red Hat system, you then create the bond0 configuration file "/etc/sysconfig/network-scripts/ifcfg-bond0" with:

Next, you edit the slave interface files. For eth0, you create or modify "ifcfg-eth0" to contain:

You do the same for eth1. You then restart the network service with "systemctl restart network" (or "ifdown eth0; ifdown eth1; ifup bond0" on older systems). Once the bond is up, you verify with "ip link show bond0". You should see that the bond0 interface has the MAC address of the first slave and is in "UP" state. You also check the bonding status by reading "/proc/net/bonding/bond0". This pseudo-file shows the active slaves, the mode, and the link status of each slave. It is the first place you look when troubleshooting.

Now test fault tolerance. While a colleague pings the server from their workstation, you unplug eth0. The ping continues with one or two lost packets — the bond switches to eth1. You plug eth0 back in and the bond re-adds it automatically (depending on the mode and miimon). The website remains available. The boss is happy because the server now has redundant network paths and slightly better throughput.

Later, the company grows and buys a managed switch that supports 802.3ad LACP. You upgrade the bond to mode 4 for better performance and standardised load balancing. You reconfigure the switch ports to be in a LACP trunk group. You update the bonding configuration on the server to "mode=4 lacp_rate=fast miimon=100". You also configure the "xmit_hash_policy" parameter to control how packets are distributed; the default "layer2" hash is fine for basic setups, but you might switch to "layer3+4" for better distribution across many connections.

As an IT professional, you will use bonding in virtualisation environments (combining NICs from the host to the hypervisor), storage networks (bonding iSCSI connections), and any production server that cannot afford downtime. The tools you use daily include: