@mowgli @dan This was in part because the bps for packetised MPEG-4 could really pop if one of our users suddenly had a proper HD stream (still somewhat rare at that point), and that presented a challenge to smooth playback in a player with the wrong concurrency design, buffer design, or drivers (which usually have their own buffers). Custom C++ memory allocators, a la Alexandrescu's work, were part of the repertoire of stuff that helped.