PPPoE introduces an 8 byte overhead per packet. The "MTU of the Internet" is 1500, so that's what more or less everything defaults to.

This includes physical NICs on Linux, but the PPPoE interface has to tunnel through one of such physical NICs.

If the physical NIC has an MTU of 1500 (and can't be changed), the PPPoE NIC must do MSS clamping, effectively reducing the MTU from my network to the Internet to 1492. This increases fragmentation and overhead.

If I can increase the physical NIC's MTU to 1508 (and the ISP supports it, which mine does), then the PPPoE tunnel can use the full 1500 when talking to the Internet.

So, it's technically not _required_ but it's an improvement I should be able to implement easily (in OpenWRT I literally type 1508 on the MTU box for the NIC, or issue a single uci command).