P2P? delivery

  • Integration with quality measurement
  • Current p2p implimentations
    • RTP
  • Comparison of different approaches (and vendors)
    • Experience measurement (loading time, buffering)
    • In a traditional CDN, you get the data as fast as possible. In p2p, you optimize for a few different things. (Time for download isn't a fair measurement)
    • Peer efficiency
    • In geographies where POPs? are far apart, you measure experience
    • For ABR, can we deliver better quality to clients that don't have strong connections to the original server?
      • ABR algorithms need to take peer connection information into account
      • Expose speed metric to the player, mimicking how http would work
  • Why p2p distribution isn't more popular
    • A lot of effort trying to deliver a very specific case
    • Academic researches came out in 2000, network capabilities were not as good at the time
    • Needed a client plugin
    • p2p can have too much latency
    • In China, there are 100 different p2p CDNs?
    • Use micropayments to solve drop off issues
    • Average participants should not be worse off
  • Should browsers have native p2p support?
  • Interop between different types of p2p
    • Some content providers specify their own protocol
  • It's difficult to debug in a p2p environment
  • media server optimization for p2p
    • resize chunks on the fly
    • reducing the segment size helps peer efficiency
    • adding http headers
    • add swarm data to the manifest
    • having large chunks at the beginning to start the session can help with efficiency
    • ABR for p2p
  • Different approaches to creating Swarms
    • Using GeoIP?, promote leader election
    • Kademlia
    • Discovery on centralized servers
  • Network topology