Can I bond interfaces from different network cards or different speeds?

Technically yes, but for LACP (mode 4) the 802.3ad standard requires all ports in an aggregation group to run at the same speed and duplex. Mixing speeds will cause one port to be excluded from the aggregator by the switch. Active-backup (mode 1) has no such restriction.

Does LACP actually double my bandwidth for a single connection?

No. LACP distributes flows across links, not individual packets (to avoid reordering). A single TCP connection between two hosts will use exactly one link. You gain aggregate throughput only when there are multiple simultaneous flows.

What is the difference between miimon and arp_interval for link detection?

miimon polls the NIC's MII (hardware) register to detect carrier loss—it is fast, reliable, and does not depend on network reachability. arp_interval sends ARP probes and can detect layer-3 path failures that miimon misses, but is slower and requires a valid ARP target. Use miimon for most cases; combine both only if you need end-to-end path validation.

My switch shows the port-channel as 'suspended'. What is wrong?

A suspended port-channel almost always means an LACP parameter mismatch: check that both sides use the same port priority, system priority, and that the switch port is set to 'mode active' or 'mode passive'—never 'mode on' (static) when the Linux side is in 802.3ad mode.

Can I add a third NIC to an existing bond without downtime?

Yes. Just add a new slave connection pointing to the existing bond master. In LACP mode the switch must also add that port to the port-channel. Traffic continues on the existing slaves while the new one negotiates LACP and joins the aggregator.

Linux Network Bonding (LACP and active-backup)

Network bonding combines two or more physical interfaces into a single logical interface, giving you either increased throughput, link redundancy, or both. Linux implements bonding through the bonding kernel module and exposes it via NetworkManager, systemd-networkd, or manual ip commands. This guide covers the two most operationally useful modes: 802.3ad LACP (mode 4, requires a cooperating switch) and active-backup (mode 1, works with any switch or even a dumb hub). You will configure the bond, verify it, and deliberately break links to confirm failover behaves as expected.

Bonding Modes Reference

Linux supports seven bonding modes. Know at least these four before you touch production:

Mode	Name	Use case	Switch requirement
0	balance-rr	Throughput, no fault tolerance guarantee	None, but can cause packet reordering
1	active-backup	Pure redundancy	None – works everywhere
2	balance-xor	Throughput + basic redundancy	Static LAG
4	802.3ad (LACP)	Throughput + redundancy	LACP-enabled switch port
6	balance-alb	Adaptive load balancing	None, but ARP-dependent

For most new deployments choose mode 4 (LACP) when you control the switch, or mode 1 (active-backup) when you do not. The rest of this guide configures both.

Switch-Side Configuration (LACP Only)

LACP (mode 4) requires both ends to speak IEEE 802.3ad. If your switch ports are not configured for LACP, use mode 1 instead; a misconfigured LACP bond will drop all traffic.

Cisco IOS / IOS-XE example

interface GigabitEthernet0/1
  channel-group 1 mode active
interface GigabitEthernet0/2
  channel-group 1 mode active
interface Port-channel1
  description bond0-server
  switchport mode access
  switchport access vlan 100

Use mode active on the switch when the Linux side also sends LACP PDUs (it does by default). mode passive on one end is acceptable but mode on (static) will not negotiate and the bond will not come up in LACP mode.

Juniper EX example

set interfaces ae0 aggregated-ether-options lacp active
set interfaces ge-0/0/0 ether-options 802.3ad ae0
set interfaces ge-0/0/1 ether-options 802.3ad ae0

Install and Load the Bonding Module

The bonding module ships with every major distro kernel; you just need to load it and, on some distros, install the userspace tools.

Debian / Ubuntu

sudo apt install ifenslave

Fedora / RHEL / Rocky

sudo dnf install NetworkManager-team

Arch Linux

No extra package needed; the module is in linux. Load it now to verify:

sudo modprobe bonding
lsmod | grep bonding

Configure with NetworkManager (Recommended)

NetworkManager is the default on Fedora, RHEL, Rocky, Ubuntu Desktop, and most modern installs. It handles bonding cleanly without manual file editing.

Step 1 – Identify your physical interfaces

ip link show

Note the exact names—commonly enp3s0 and enp4s0 or similar. Substitute them in every command below.

Step 2 – Create the bond master (LACP)

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

For active-backup instead, change mode=802.3ad to mode=active-backup and drop the lacp_rate and xmit_hash_policy options; they are irrelevant for mode 1.

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

fail_over_mac=active avoids MAC conflicts when connected to switches that track port MACs strictly.

Step 3 – Add slave interfaces

sudo nmcli connection add type ethernet \
  con-name bond0-slave1 \
  ifname enp3s0 \
  master bond0

sudo nmcli connection add type ethernet \
  con-name bond0-slave2 \
  ifname enp4s0 \
  master bond0

Step 4 – Assign an IP address to the bond

sudo nmcli connection modify bond0 \
  ipv4.addresses 192.168.1.50/24 \
  ipv4.gateway 192.168.1.1 \
  ipv4.dns "1.1.1.1 8.8.8.8" \
  ipv4.method manual

Step 5 – Bring it up

sudo nmcli connection up bond0
sudo nmcli connection up bond0-slave1
sudo nmcli connection up bond0-slave2

Configure with systemd-networkd (Servers / Headless)

On Ubuntu Server with systemd-networkd, or on Arch with networkd enabled, drop these files into /etc/systemd/network/.

Bond netdev

sudo tee /etc/systemd/network/10-bond0.netdev <<'EOF'
[NetDev]
Name=bond0
Kind=bond

[Bond]
Mode=802.3ad
TransmitHashPolicy=layer3+4
MIIMonitorSec=100ms
LACPTransmitRate=fast
EOF

Bond network (IP assignment)

sudo tee /etc/systemd/network/10-bond0.network <<'EOF'
[Match]
Name=bond0

[Network]
Address=192.168.1.50/24
Gateway=192.168.1.1
DNS=1.1.1.1
EOF

Enslave each interface

for iface in enp3s0 enp4s0; do
sudo tee /etc/systemd/network/20-${iface}.network <<EOF
[Match]
Name=${iface}

[Network]
Bond=bond0
EOF
done

sudo systemctl restart systemd-networkd
sudo systemctl enable systemd-networkd

Verifying the Bond

Always verify before relying on a bond in production.

Read the kernel bond status file

cat /proc/net/bonding/bond0

You should see MII Status: up for each slave, the active slave (mode 1) or aggregator ID (mode 4), and the bonding mode name. For LACP, confirm 802.3ad info is present and each port shows LACP Activity: Active.

Check link state with ip

ip link show bond0
ip -s link show bond0

Confirm connectivity

ping -I bond0 192.168.1.1 -c 4

Fault-Injection Testing

A bond you have never broken is a bond you cannot trust. Test failover deliberately.

Method 1 – Physically unplug a cable

Unplug one patch cable and watch /proc/net/bonding/bond0. For mode 1 you will see the active slave change within miimon milliseconds. For mode 4 the aggregator will drop one port but traffic continues on the remaining link.

Method 2 – Bring a slave down in software

# Disable slave 1
sudo ip link set enp3s0 down

# Watch bond status in real time
watch -n1 cat /proc/net/bonding/bond0

# Re-enable
sudo ip link set enp3s0 up

Method 3 – ethtool carrier injection

# Simulate carrier loss (requires ethtool and a driver that supports it)
sudo ethtool --set-priv-flags enp3s0 disable-fw-lldp on
sudo ip link set enp3s0 carrier off 2>/dev/null || echo "Driver does not support carrier injection"

Not all drivers support carrier injection; falling a slave down via ip link set down is the most portable approach.

Measure failover time with ping

ping -I bond0 192.168.1.1 -i 0.2 &
sudo ip link set enp3s0 down
# Count dropped pings, then restore
sudo ip link set enp3s0 up

With miimon=100 expect 1–3 dropped pings during failover. Tuning miimon lower (e.g. 50ms) reduces the window at the cost of more kernel polling overhead.

Monitoring in Production

systemd service to alert on slave state change

The kernel emits uevents when bond slave state changes. You can hook into them via a udev rule:

sudo tee /etc/udev/rules.d/90-bond-alert.rules <<'EOF'
SUBSYSTEM=="net", ACTION=="change", ENV{MASTER}=="bond0", \
  RUN+="/usr/local/bin/bond-alert.sh %k"
EOF

Have bond-alert.sh send a notification via your preferred channel (email, Slack webhook, PagerDuty). Reload udev after saving:

sudo udevadm control --reload-rules

Check LACP PDU counters

cat /proc/net/bonding/bond0 | grep -A5 "Slave Interface"

Healthy LACP PDU exchange shows incrementing counters. Stalled counters indicate the switch stopped sending LACP PDUs—check your switch configuration first.

Troubleshooting

Bond stays in mode 4 but traffic drops entirely: The switch ports are not configured for LACP. Switch to active-backup temporarily to restore connectivity, then fix the switch config.
One slave never joins the aggregator (LACP): Run cat /proc/net/bonding/bond0 and check the aggregator ID. A slave with a different aggregator ID usually means a switch port speed/duplex mismatch or LACP PDU not reaching the switch. Check ethtool enp3s0 for speed negotiation.
NetworkManager keeps disconnecting slaves: Ensure each slave connection has connection.autoconnect-slaves 1 set: sudo nmcli connection modify bond0 connection.autoconnect-slaves 1
MAC address flapping on switch: In active-backup mode, set fail_over_mac=active to prevent both slaves from presenting the same MAC simultaneously during failover.
Bonding module options not persisting after reboot: With systemd-networkd, options live in the .netdev file. With NetworkManager they persist in the connection profile. Do not also write to /etc/modprobe.d/bonding.conf or you may get conflicting parameters.