improve latency

[slugalisk]

it's not clear what is contributing the most to this. latency varies from 10 to 20 seconds for peers connected directly to the stream source...

suspects...

• ffmpeg/obs
  • probably negligible?
  • are there encoder settings to optimize for this?
• nginx-rtmp module
  • this currently splits on keyframes... could we generate smaller chunks and flush more frequently if we ignore keyframes? the chunks are given as a continuous stream to mux.js anyway...
  • is mpeg ts the right format for this? does something like rtmp/rtsp make more sense?
• chokadir
  • is a 1 second stabilityThreshold overly conservative?
• chunked read/write stream
  • should we zero pad the input chunks to align with the injector chunk size?
  • does the spec provide an efficient way to encode empty space?
• injector
  • there are several settings here that could be adjusted...
• scheduler
  • the first available/needed chunk to pass a random roll with a 5% chance is currently picked... this strategy probably results in waiting too long to prioritize lagging chunks. the spec describes a priority and reputation based system for assigning requests for urgently needed chunks. this should be implemented.
• player
  • could we be flushing the transmuxer more frequently?