How to change mtu size?

[Unit]
Description=Change MTU Service
After=connman.service
Before=network-online.target

[Service]
Type=oneshot
StandardOutput=tty
RemainAfterExit=yes
SuccessExitStatus=1
ExecStart=/usr/sbin/ifconfig eth0 mtu 9000 up

[Install]
WantedBy=network-online.target

Create /storage/.config/system.d/mtuchange.service with ^ then "systemctl enable /storage/.confg/mtuchange.service" and "systemctl start mtuchange.service" .. it's a guess (not tested) but should run after connman starts and before the network goes online.

autostart.sh runs at the start of userspace boot long before networking is up, so the only way to sequence things is backgrounding with a sleep delay and then hoping the timing remains consistent over time. Using a systemd service allows proper sequencing to remove all the guesswork on timing.

Quote from chewitt

Create /storage/.config/system.d/mtuchange.service with ^ then "systemctl enable /storage/.confg/mtuchange.service" and "systemctl start mtuchange.service"

Just to be sure: I guess, that should be

Create /storage/.config/system.d/mtu.change.service

and then copy & paste your code lines and save it

and then

systemctl enable /storage/.config/system.d/mtu.change.service

followed by

systemctl start mtu.change.service

Unfortunately, my Rasperry Pi tells me at the boot process and executing the above

/usr/sbin/ifconfig eth0 mtu 9000 up

ifconfig: SIOCSIFMTU: Invalid argument

How should the syntax be set to get the MTU working with a value 9000 ?

The maximum MTU size supported on RPi boards is 1500, hence the "Invalid argument" error.

See https://github.com/raspberrypi/linux/issues/5561 and https://github.com/raspberrypi/linux/pull/5534 and an earlier refused PR trying to add the same (not something I'd recommend trying) https://github.com/raspberrypi/linux/pull/5419.

It might be possible to get Jumbo frames with an external USB-C NIC, but I couldn't recommend one, and IMHO there's no real need for Jumbo frames on RPi hardware. RPi 0/1/2/3 are too slow to really benefit and RPi4/5 are already fast enough that the benefit will be negligible under real-world usage (the only real difference will be under un-real lab tests).