Article: Resource and Locality Awareness in an Incentive-based P2P Live Streaming system

Unlike VoD, playback delay of a live stream monotonically increases without explicit bounds => until data too outdated to be available any longer.

They use a P2P system called "PULSE".

In a live P2P system, a node has a Trading Window: it is only interested in chunks from this window (not previous chunks, but not too far ahead either).

Can a remote node provide needed chunks ? Interest in a remote node:

• proportional with the overlapping of the Trading Windows
• decreases with link latency (latency reduces the window from which one can request chunks from a peer)

Simulation results:

1. direct relationship between upload bandwitdh and average playback lag (for a whole class of peers): high upload bw rewarded with small lag = resource awareness
   
2. data paths: trees describing the path of a chunk through the network: good, stable tree depth, good tree width in first layers, for high and low bandwidth
   
3. when introducing latency cost in peer selection, peers clearly begin exchanging more data locally (with closer peers) = locality awareness
4. when introducig latency cost in peer selection, peers get an average reduced playback lag, because they are forced into exchanging data with physically closer peers (small reduction, but they expect larger reduction in a scenario with large possible latencies between peers).
   

Conclusion:

• the paper shows by way of simulations how an incentive-based peer selection algorithm supports heterogenous resource availabilty in a distributed system
• in my words, how the algorithm of selecting peers influences the system: latency cost => forces local peer exchange, favouring peers which are more useful (upload more to you, BitTorrent style) => small lag for high upload bandwidth