Get training via VPN to work

[oleschwen]

We're currently observing that trainings are aborted in varying rounds if nodes communicate via VPN. This might be a bandwidth/timeout issue.

AC: DUKE training finishes successfully (at least twice) with

• two clients at one site communicating directly
• two clients at one site communicating via local test VPN (bandwidth limited to 30? Mbit/s, similar to production VPN)
• two clients at one site communicating via production VPN
• two clients at two sites communicating via production VPN
• three clients at two sites communicating via production VPN
• two clients at one site communicating via local test VPN (bandwidth limited to 4 Mbit/s, mimicking expected transfer times for 6 sites)