Linux interface configuration - addresses, MTU, offloads · Linux

Interface configuration on Linux happens at three levels:

Logical: address, route, MTU.
Driver: ring buffers, queue count, interrupt coalescing.
Hardware / firmware: offloads, RSS, firmware settings.

The right tool depends on the level. ip works at level 1; ethtool works at levels 2 and 3.

Configure a static address

ip addr add 10.0.0.10/24 dev eth0
ip addr add 2001:db8:85a3:1::10/64 dev eth0
ip link set eth0 up

For a working host, you also need a route:

ip route add default via 10.0.0.1

And, for IPv6, a default route via the link-local router:

ip -6 route add default via fe80::1 dev eth0

These are all transient. They are useful for debugging but disappear at reboot. Persistent configuration is covered in the Netplan, NetworkManager, and systemd-networkd lessons.

MTU

MTU (Maximum Transmission Unit) is the largest IP packet an interface will send without fragmenting. The default on Ethernet is 1500 bytes. Jumbo frames use 9000; some tunnels require 1400-1450.

ip link show eth0                # current MTU
ip link set eth0 mtu 1400        # change it
ip link set eth0 mtu 9000        # jumbo frames

A wrong MTU causes silent packet loss. Symptoms:

Small flows work; large transfers fail.
TCP connections stall after handshake.
ping -M do -s 1472 fails; ping -M do -s 1400 works.

Offloads

Network cards can do work in hardware to relieve the CPU:

Offload	What it does	Trade-off
TSO (TCP Segmentation Offload)	NIC splits large TCP segments	Saves CPU; can confuse some packet captures
GRO (Generic Receive Offload)	Coalesces incoming packets	Reduces CPU; can hurt latency measurements
LRO (Large Receive Offload)	Older, similar to GRO	Deprecated in favour of GRO
RSS (Receive Side Scaling)	Hashes flows to queues	Needed for multi-queue performance
checksum offload	NIC computes IP/TCP/UDP checksum	Saves CPU; needed for high throughput
scatter-gather	DMA from multiple buffers	Saves CPU

ethtool -k eth0                  # show all offloads
ethtool -K eth0 tso off          # disable TSO
ethtool -K eth0 gro off          # disable GRO

Common reason to disable offloads: packet captures look wrong, or an application measures sub-microsecond latency and the coalescing adds jitter.

Ring buffers

The NIC ring buffer holds packets between the NIC and the kernel:

ethtool -g eth0                  # show current and max
ethtool -G eth0 rx 4096          # set the rx ring to 4096

A small ring buffer drops packets under load (visible in ifconfig errors or ip -s link). Increase the ring if you see drops and CPU has headroom.

Tuning low-level driver parameters

ethtool -c eth0                  # show coalesce settings
ethtool -C eth0 adaptive-rx on   # adaptive rx coalescing
ethtool --set-priv-flags eth0 <flag> on   # driver-specific

Driver-specific settings vary. Always read the driver’s documentation before changing private flags.

Verify configuration took effect

ip -br addr show eth0
ip -br link show eth0
ip route show dev eth0
ethtool eth0
ethtool -k eth0 | grep -E 'tcp-segmentation|generic-receive'

If ip shows the change but the application still misbehaves, restart the application or the network service. Some applications cache the MTU at startup.