[mosh-users] status of UDP hole-punching?

[Thomas Sattler]

Hello Keith

> (I assume we're talking here about having the _server_ be
> behind a NAT. Having the client behind a NAT works fine.)

Actually I'm talking about both ends being behind NAT: client
*and* server. I'm using mosh for several months now, and all
but two servers work fine. These two are behind gateways
where I cannot add forwarding rules.

> Somebody has suggested that for full-cone NATs, the server
> really just needs to punch through _once_

I'm not sure if this is correct, might also depend on the
NAT gateway in use. But that's far beyond my knowledge.

How about Philipp's approach where the client can initiate
"repunching" if needed. He needs to login via ssh again to
trigger this, but hey, having this is better than nothing.

Aren't there plans to implement something to reconnect to
a lost server after (a client side) crash or reboot? Seems
not to be too different.

> My best recommendation for now is to do whatever you do for
> SSH. If you forward port 22 to your server for SSH, you will
> need to forward UDP ports 60000-(something higher) for Mosh
> as well.

I did this on n-2 machines, looking forward to eventually
be able to use 'mosh' on all my machines.

Thomas