OpenVPN slow without sndbuf 0 / rcvbuf 0

[hjohnson57]

I'm running OpenVPN 2.4.0-6+deb9u2 on my Debian 9 server and OpenVPN 2.4.6-I601 on my Windows 10 client. The issue is not related to a specific version of either of these. RTT latency between these is about 40msec (this is relevant).

openvpn-install.sh works great, but is slow. Installed as tcp-over-tcp with no packet loss (it's faster that way). By default, I get about 5Mbps throughput and links on both ends are at least 600Mbps up/down in both directions (this is also relevant). If I add the following to the scripts on both ends, I get about 60Mbps each way:

sndbuf 0
rcvbuf 0

Feature request: include these two lines in openvpn-install.sh.

See https://www.lowendtalk.com/discussion/40099/why-openvpn-is-so-slow-cool-story for details.

[angristan]

Thanks for the link. What's weird though:

As you already may guess, the latest OpenVPN release still uses 64 KB socket buffer size. How should we fix this issue? The best way is to disallow OpenVPN to set custom buffer sizes. You should add the following code in both server and client config files:

sndbuf 0
rcvbuf 0

After this, socket buffers would be handled by OS.

But from the manual:

--sndbuf size
Set the TCP/UDP socket send buffer size. Defaults to operation system default.
--rcvbuf size
Set the TCP/UDP socket receive buffer size. Defaults to operation system default.

[hjohnson57]

I'm guessing that the manual is out of date. Also, there are separate send and receive buffers for each end of the VPN connection.
Since my last post, I've also noticed the buffer bloat https://www.bufferbloat.net effect with this setting on OpenVPN over TCP (but not over UDP). So this setting may exhibit unwanted latency over TCP.