I just returned from the GStreamer Conference
at Edinburgh and I'm still processing the extremely interesting conversations we had there. Once again the conference was a great success, fascinating and great fun. Big kudos to the awesome organizers.
After a quick welcome by Christian Schaller, Tim-Phillip Muller gave the keynote. His approach was to get the ball rolling and get the conversation going from the beginning, instead of giving the usual summary of what has happened in the last year he looked forward into the future and talked about taking GStreamer to the next level. Not only patting ourselves in the back for a good year of development but also reminded us of what is missing and should be done. Afterwards he asked the audience and there was heavy emphasis in more and better documentation and testing from the participants.
Thanks to Tim the hall conversation where very constructive from the get-go, but after the coffee break I went back to the Kylsith room to be amazed by David Röthlisberger's automated testing of set-top boxes user interfaces. He introduced us to stb tester
, which is occam's razor applied to testing. With the power of OpenCV and GStreamer, they have built a Python state machine that navigates the menus of the set-top box by triggering infrared signals, recognizes patterns/behaviour on the screen and checks any use case you might consider.
Right after came David Schleef, who was been working on GStreamer Streaming Server
. Solving the common and interesting modern web problem of smooth streaming, via bitrate switching and adaptive streaming. Thanks to his work you can easily stream live video streams to thousands of clients from an HTTP server. I definitely plan to play with this technology in the near future.
My two favorite talks were yet to happen. Tuesday brought us Edward Hervey talking about Time \ and Synchronization. I appreciate that he took the time, no pun intended, to explain the concept of time itself before the actual problem of synchronizing with it. A refreshing summary about how time isn't a thing or a unit, but durations of time are what we use. How clocks work and how sometimes there is no reference on when they started and their pace, which complicate things. Ending with the three times GStreamer has and why. This topic is too long and enjoyable to do it any justice by summarizing it, I will update once Ubicast
has the talk videos out. I recommend watching it.
Wednesday had the surprise of the conference. Håvard Graff presented gst-harness. A deterministic testing infrastructure for GStreamer that can properly test elements in a self contained sandbox. They are planning to merge this into gst-tools before the end of the calendar year. Of everything I saw and heard during the conference, this is without a doubt, the thing that could change the GStreamer ecosystem the most. Huge potential for improving and solidifying the stability of the elements.
All in all, the GStreamer Conference is still my favorite conference of the year. The content of the talks is just superb, and tied with GUADEC with how cool the attendants are. Big hug or chest bump to everybody I talked with there, and a big thank you to the organization and sponsors. See you all next year!