networking

Making the Intel X710-DA4 Work…

Posted on by dk

Apparently, making this 4x SFP+ card work is not as straight forwards as I thought…

  • LACP (aka Link Aggregation aka 802.3ad) would not work properly under Linux, Windows, VMware or FreeBSD unless a “private flag” was disabled using ethtool (source #1, #2)

Sigh, so much for “plug and play pray”…

As per the source/s instructions, I created a /etc/networkd-dispatcher/configuring.d/90-disable-on-chip-LLDP.sh:

#!/bin/sh
if expr "$IFACE" : <interface prefix> >/dev/null; then /sbin/ethtool --set-priv-flags "$IFACE" disable-fw-lldp on fi

WARNING: Make sure your <interface prefix> is able to only address the X710 interfaces, or you will likely get errors or, worse, brick your NIC. As an example, my X710 interfaces are all named enp1s0fN so I use the <interface prefix> of “enp1s0f”

To check, simply give the command ethtool --show-priv-flags <interface prefix>.

netplan Not Going To Plan…

Posted on by dk

With a /etc/netplan/00-global.yaml having the following contents:

network:
  version: 2
  renderer: networkd

And all other .yaml files in the /etc/netplan directory not having an explicit renderer, the system should default to networkd… Just netplan apply and done! Simple, right?

Yeah, right.

Seems like netplan apply still thinks to use NetworkManager even when no single .yaml file declares its use; and since I disabled and masked the service, netplan apply was complaining NetworkManager was masked.

The secret sauce?

Continue reading →

The Qotom C3758R Unit Continued…

Posted on by dk

Just a quick update on networking experiences with this fanless (but hot) unit…

 

Why Is It Not In Sync? (buy buy buy)…

For some odd reason, throughput on stock Ubuntu 24.04 LTS via a 10GbE SFP+ module was showing asymmetrical speeds on iperf3 – ~10Gbps in, but only ~6Gbps out. Checking the firewall, ensuring jumbo packets (9000 byte MTU), ensuring ufw was disabled, disabling all forms of NIC offloading (although this may actually be inverse to the intended) did not change anything. Furthermore, htop showed one core constantly maxing out when sending (but hovering 60% to 75% on receiving).

Continue reading →