When troubleshooting a network problem, the first task is confirming what speed the NIC actually negotiated — a lit link light does not mean the link is running at its expected bandwidth. A gigabit port negotiating down to 100M, or auto-negotiation failing and downshifting the link: these are not rare in a server room. Linux offers three common entry points for NIC information: ethtool for link-layer speed, ifconfig (or its modern replacement ip addr) for interface state, and nmcli for device details under NetworkManager. This post is organized by “what you want to see” — find your row and read.

All output in this post was verified on a Rocky Linux 9.3 container (ethtool 6.15, net-tools 2.0, iproute 6.17, NetworkManager 1.54.3, aarch64). On CentOS 7/8/9 and the RHEL family the commands and fields are the same; the numbers vary by environment.

For Negotiated Speed and Duplex: ethtool

ethtool queries the link-layer state straight from the NIC driver — the first choice for speed. A Rocky Linux 9 minimal install does not ship it; install ethtool and net-tools together, since both appear in this post:

1
dnf install -y ethtool net-tools

Figure 1

Note: Rocky 9’s dnf is usage-compatible with CentOS 7’s yum; swapping dnf for yum in these commands works on CentOS 7 too.

Then query a specific interface (eth0 throughout this post; substitute your actual interface name):

1
ethtool eth0

Figure 2

The full output is long; four fields matter first when troubleshooting:

  • Speed: the currently negotiated rate — 10000Mb/s means a 10-gigabit link. A gigabit NIC showing 100Mb/s: check the cable and the peer port first, not the OS.
  • Duplex: the duplex mode. Full is full duplex; Half invites collisions and dropped packets on the link, usually the result of a failed auto-negotiation downgrade.
  • Auto-negotiation: whether auto-negotiation is on. Both ends must be configured consistently; one end fixed and the other negotiating is the most common configuration trap.
  • Link detected: whether the link is up. When it shows no, swap the cable first, then check the switch port.

Note: the Figure 2 output carries one extra line at the end, netlink error: Operation not permitted, caused by the container lacking privileged bits. It does not appear on physical machines and does not affect reading the fields.

When installing drivers or matching a NIC model, add ethtool -i eth0; the driver field names the driver directly — veth in the container, and on physical machines commonly igb, e1000, ixgbe, or virtio_net:

Figure 3

Beyond querying, ethtool can also change negotiation parameters temporarily (ethtool -s eth0 speed 1000 duplex full), but for troubleshooting the read-only query is what you use most — changing parameters is a change operation that needs a change record, it is lost on reboot, and making it permanent requires the connection configuration (see the caveats at the end).

For a Quick Interface State Check: ifconfig

ifconfig answers “is this interface up, is the address right”: IP, netmask, broadcast address, MTU, plus packet counters — all in one view:

1
ifconfig eth0

Figure 4

In the eth0 line’s flags, UP,RUNNING means the interface is enabled and the link is healthy; the inet line is the IPv4 address and mask; RX/TX packets are the receive and transmit counters. If RX errors keeps growing, something is wrong at the link layer — go back to ethtool for speed and duplex.

Two reminders: ifconfig belongs to the net-tools package, absent from minimal installs from CentOS 7 through Rocky 9 — the dnf at the top of this post installed it along the way; and the package itself has not been updated in years, so on newer systems prefer the iproute2 equivalents — ip addr for addresses, ip -s link for counters, whose output conventions match ifconfig’s:

1
ip -s link show eth0

Figure 5

The Full Device Picture Under NetworkManager: nmcli

Since the RHEL 7 family, the network is managed by NetworkManager by default, and nmcli is its command-line entrance. nmcli device show lists a device’s type, driver, MAC, and full IPv4 configuration:

1
nmcli device show eth0

This command requires the NetworkManager daemon to be running. The container had the NetworkManager package installed (1.54.3) but no daemon, and the command failed directly, as verified:

Figure 6

On a normal server (NM running), the output is longer than the previous two commands and suits a one-time full device inventory. Per the nmcli(1) manual, GENERAL.DRIVER is the NIC driver name (matching the driver from ethtool -i) — useful when installing drivers or tuning parameters; IP4.ADDRESS[1] is the currently effective address, faster than digging through config files. It also has one irreplaceable use: after modifying a connection configuration (nmcli con ...), re-run the same command to check that configuration and actual state agree.

Command Cheat Sheet

Command What it shows Typical scenario
ethtool eth0 Speed, duplex, auto-negotiation, link Bandwidth below expectation, packet-loss investigation
ethtool -i eth0 Driver name, firmware version Matching a NIC model, confirming before driver install
ifconfig eth0 IP, netmask, MTU, RX/TX counters Quick interface state and address check
ip addr / ip -s link Addresses and interface state (with counters) Replacing ifconfig on newer systems
nmcli device show eth0 Driver, MAC, full IPv4 configuration Device inventory under NetworkManager

Historical Version Differences (CentOS 7)

For readers coming from CentOS 7, the three commands themselves are unchanged; the differences are in the surroundings:

  • Install command: yum → dnf, usage-compatible and a direct swap; CentOS 7 is EOL and its official repos are offline — you must switch to the vault repos to install packages.
  • NIC naming: consistent naming has been the norm since CentOS 7 (ens192, ens33 and the like) rather than eth0 — run ip link to list the actual interface names first, then substitute.
  • net-tools absent: CentOS 7 minimal installs also lack ifconfig — old and new systems agree there; the difference is that on newer systems the officially recommended slot has gone to iproute2.

Caveats

  • Substitute your environment’s actual interface name. RHEL 8/9 occasionally retains custom naming rules; go by the actual output of ip link, not a copied eth0.
  • Suspect the physical layer first when speed is wrong. Low Speed or Duplex showing Half: check the cable, optical module, and peer switch port before anything at the OS level.
  • A fixed speed only counts once it lands in configuration. ethtool -s is temporary and lost on reboot; persistence goes into the connection configuration (nmcli’s ethtool-related settings), and both ends must be fixed simultaneously — fixing only one end leaves the link either unable to come up or downshifted.
  • For scripts, prefer ip -4 addr show eth0 with grep to extract IPs — steadier than depending on ifconfig.
  • Testing NIC parameters inside a container means little: veth speed/duplex are virtual values handed down by the host; capacity planning should be based on physical machine measurements.

Summary

For bandwidth troubleshooting: use ethtool to confirm what the link negotiated (-i adds driver information), then ifconfig or ip -s link to confirm interfaces and addresses, and in a NetworkManager environment nmcli gives a device’s complete profile. Three entry points, each covering one layer — pick by what you want to see.

This article was rebuilt from the author’s CSDN blog posts written between 2020 and 2024, originally published on CSDN.