for work we rent a dedicated server from OVH. it’s been 5 months now and i’m pretty satisfied with the service provided. at the initial stage we’ve bumped into a problem that was never really solved; i cannot be even 100% sure if it’s OVH’s fault.
UDP-based OpenVPN connection established from OVH’s BHS datacenter to IP addresses in Poland and Sweden [Bahnhof] had terrible performance and packet losses – it looked to me like the encrypted connection was throttled to ~2mbit/s while we could easily push 60-90mbit/s over tcp or udp. i exchanged some mails with the OVH’s support – but – to my despair – they denied any throttling. the problem remains unsolved but i found a simple workaround – i have a GRE tunnel established from OVH to both endpoints and i run OpenVPN via GRE. it’s not a very elegant solution, but it works just fine and we’re pushing at peak 70-90mbit/s each day via that encrypted tunnel.
on one end of the tunnel [machine with ip: 192.95.ovh.ip] i have:
ip tunnel add tun20 mode gre local 192.95.ovh.ip remote 79.136.ip.inSweden ttl 225 ip link set dev tun20 up ip address add dev tun20 10.1.1.4 peer 10.1.1.3/32
on another end [linux box with ip 79.136.ip.inSweden]:
ip tunnel add tun20 mode gre local 79.136.ip.inSweden remote 192.95.ovh.ip ttl 225 ip link set dev tun20 up ip address add dev tun20 10.1.1.3 peer 10.1.1.4/32
and openvpn is configured to use 10.1.1.3, 10.1.1.4 addresses rather than public ips. i lose a bit of bandwidth on encapsulation but still get 60-80mbit/s via such vpn instead of mere 2mbit/s.
2015-11 edit: we’ve been using UDP over IPv6 as an openvpn transport sine ~ April 2015 and it gave us most stable connection so far.