Firstly, it's not fragile, because it has good state recovery. The old PPTP VPN would, if we moved network or suspended and awoke our laptops, go down and require manual re-connection. However, n2n works by having the edge nodes send regular keep-alives to the supernodes, updating the supernodes' directory tables. So if we're offline, the edge nodes' keep-alives are discarded due to lacking a route out; and when we come online again, one gets to the supernode, and then we're back.

This means that we've ended up leaving our n2n connections running, as once they're running, they stay running. Being connected to the VPN is something that's just there, rather than being something you manually initiate when needed.

And this has led to a second dimension: the laptops of anybody who's online are all joined on a virtual Ethernet LAN. So we can see if people are online by just pinging them. And we see each other appear in Bonjour.

We have a Nabaztag in the office; it announces (via text-to-speech) when people push to 'mainline' git repositories, and we want to make the continuous-integration system so it can announce when somebody breaks the build. But that's not so useful for people who are working from home; so we've been considering running a reliable-multicast group (using the Spread Toolkit) that team events can be published into, and things like the Nabaztag listen from - but then we can extend it over our VPN, so it can drive Growl notifications on my Mac, and similar mechanisms for the Linux users.