Recent discussion: <a href="https://news.ycombinator.com/item?id=10116879" rel="nofollow">https://news.ycombinator.com/item?id=10116879</a><p>Same guy being interviewed in a software-defined-networking podcast<p><a href="http://blog.ipspace.net/2015/03/tcp-optimization-with-juho-snellman-on.html" rel="nofollow">http://blog.ipspace.net/2015/03/tcp-optimization-with-juho-s...</a>
I got involved in a video streaming project for mobile last year. The back end had all sorts of aggressive dynamic bandwidth optimisation: start low, switch to a higher quality if you can and so on. I told the dev team they were making assumptions about internal mobile network topology that they couldn't possibly know and that it would bite them. It did. So badly in fact that there was a dispute about deliverables between them and the owner of the brand. Their solution didn't work well on some carriers and was almost unusable on others. Looking back, I'm sure it was because of some of the issues raised in this presentation.<p>Over a long weekend, I cobbled together a very basic ffmpeg and nginx-rtmp-module solution using fixed low bandwidth videos (about 8k per second which works on EDGE very nicely - think Africa). It not only ran better - mobile networks don't mind giving you a small slice as long as it's predictable - but used about half a million dollars less hardware on our side.