Ah, the monastery is reachable again!

Thank you - if everything else fails, I will try this approach. It might be possible just to break all connections, re-establish them, and start over - we shall see.

But why does IO::Socket::INET ignore the read timeout setting? As far as I can see, connect timeout works, and we can see from "ordinary" Windows apps (e.g. browsers) that it is possible to time out on TCP read on Windows. Well, perhaps they too jump through hoops like this.