Re: Wireguard, MTUs, and jumbo packets

From: <freebsd-net_at_brettglass.com>
Date: Tue, 05 Apr 2022 14:43:54 UTC
Interesting. That 60 byte overhead (which I understand is intended 
to make the protocol connectionless) is concerning, because to send 
a 1500 byte packet one needs to have enabled jumbo packets along 
the entire path. Otherwise, there will be a lot of fragmentation... 
which in turn will create yet more overhead.

It also means that - at the other end of the packet size range - 
protocols that send many small packets (e.g. VOIP) could have 
tremendous overhead - 100% or more. Even TCP ACKs become 
substantial. This is a big disadvantage compared to protocols such 
as L2TP and PPTP. Is there any provision in Wireguard for packet 
aggregation? If not, there likely should be.

--Brett Glass

At 01:27 AM 4/4/2022, Patrick M. Hausen wrote:

>Hi all,
>as far as I know WireGuard does not care about interface or PMTU
>nor perform PMTUd. You can set the WG interface MTU in the configuration, e.g.
>
>         [Interface]
>         PrivateKey = **************
>         Address = [...]
>         DNS = [...]
>         MTU = 1280
>
>Wether your path will be capable of transporting packets with a tunnel MTU
>of 1500 is left for you to take care of - outside of WG.
>
>WireGuard overhead is 60 bytes for IPv4 transport and 80 bytes for IPv6.