Threshold bipolar scheduling for P2P live streaming

In P2P live streaming, the startup mechanism used by peers directly determines their streaming performance when they first join a channel or recover from playback freezes. However, there has been no systematic study of peer behaviors and streaming strategies in their startup stages. Motivated by a measurement study of peer startup behaviors in real P2P live streaming systems, we propose a simple and effective hybrid chunk fetching strategy, called Threshold Bipolar (TB), which partitions the buffer of a peer into head and tail parts and employs different chunk download strategies for the two parts. According to the TB strategy, a peer downloads chunks in the head part using selfish strategies, like the Greedy. Once the head part is fully filled, it needs to use altruistic fetching strategies, like the Rarest First or the Random, to download chunks in the tail part. We study the design choices of the major parameters of the proposed TB strategy. In particular, we propose to dynamically adjust the TB threshold to tradeoff between a peer's selfishness and altruism based on the system-wide streaming performance. We further develop analytical models to characterize the buffer progresses of peers in startup stages. Through extensive simulations, we demonstrate that the proposed TB strategy outperforms the existing P2P chunk fetching strategies and is robust against flash-crowd.