Re: bufferbloat in the tunnel
2012-06-05 11:19:08 GMT
On 05/17/2012 02:34 PM, Brian J. Murrell wrote: > I then moved the ping into the VPN and saturated the uplink inside the > VPN and was observing ping times on the order of 2000-3000ms or higher. > I then tried reducing the txqueuelen on the tun0 interface in steps > down from 100 to 50 to 10 to 5 and eventually got down to a value of 1 > on an uplink of 512Kb/s while maintaining full upstream bandwidth. At a > txqueuelen of 1, pinging inside the tunnel was a nice respectable > 20-30ms. When I stepped it up to 2, it went to 100-200ms. But clearly > since 1 still achieves full bandwidth, that is the "ideal" value. we did some testing on a server with ~ 100 concurrent connections, and we need a txqueuelen of 250 otherwise we get a lot of packet loss (and the "write UDPv4 []: No buffer space available" error). That server has a lot more bandwidth (but we only use about 10Mbps). So perhaps the default of 100 is too high for a 512Kbps uplink, it is too low for a 10Mbps uplink. Possibly the OpenVPN manual could add some useful defaults for different situations? e.g. use txqueuelen 10 for a client on a consumer ADSL connections, 100 for a 10Mbit upstream, use up to 250 for a larger site with more bandwidth? Olivier ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions(Continue reading)
RSS Feed