On Fri, 2014-09-26 at 13:33 +0100, David Woodhouse wrote: > > > (Has anyone been running VoIP over ocserv connections, btw? This talk of > > > buffers and EAGAIN reminds me that we need to make sure we avoid > > > excessive buffering. cf. > > > http://git.infradead.org/users/dwmw2/openconnect.git/commitdiff/3444f811 > > > ) > > > > I am. One can tune it using the output-buffer parameter. Since you > > brought that, I suspect that this particular commit must have been the > > responsible for the asymmetry in upload/download (it was on an old > > thread, with openconnect upload being 4 times slower than download). > > Unfortunately I have no longer the hardware to verify that theory. > Hm, maybe I was a little too aggressive about cutting down the output > buffering? But if we ask for MTU*2 don't we actually get to put *four* > full-MTU packets into the socket before it stalls? Surely that ought to > be enough to keep the packets flowing? Well VoIP packets are quite short, and may not be easy to find a value that would allow both low latency and bandwidth. If I remember well g.729 uses 20-bytes for 20-ms while g.711a 200 bytes, meaning with 4 mtu packets you can get a 500ms delay on g711a, but in g.729 it could be 5 seconds. It would be best to leave some reasonable default that does not really hamper bandwidth, and allow configuration for certain cases. I for example noticed that I use a 60-mtu packets buffer with g711a, and have not noticed jitter during talk. > Was there anything special we needed to test this? Just ocserv and > openconnect running on different hosts on a gigabit Ethernet segment? That's what I was using. regards, Nikos
