NeTS: Medium: Collaborative Research: Modeling Design and Emulation of P2P Real-Time Streaming Networks

Project: Research project

Description

Work to be Accomplished at Arizona State University The following work will be accomplished at Arizona State University. 1. Multi-Server P2P Streaming Networks To improve the quality of P2P streaming, we can deploy multiple servers in the network to push multiple chucks to multiple peers simultaneously. Consider the hybrid policy and a simple server allocation strategy such that Ki servers are used to distribute the newly generated chunk and Kt servers are used to distribute the chunk closest to playout. In this case, the probability that a peer obtains the newly generated chunk increases to Ki M , and the target probability of obtaining the chunk closest to playout from other peers can be reduced to ptarget - Kt M because a peer can successfully download the chunk from servers with a probability Kt M . Then, the minimum buffer requirement under the hybrid policy scales approximately as log M Ki + log 1 1 - ptarget + Kt M . It is easy to see that Ki and Kt should be carefully chosen to minimize the buffer requirement (or the playback delay). Assuming the simple server allocation strategy is used, the optimal server allocation can be obtained by solving the following optimization problem: min Ki+Kt=S log M Ki + log 1 1 - ptarget + Kt M , where S is the number of servers available. In reality, a server can distribute any chunk, not just the chunks farthest or closest to playout. A server can also either actively select peers and push the chunks, or passively wait for peers to pull the chunks. To that end, server allocation strategies should balance the resource allocated to push and pull. In this project we will develop insightful equilibrium models incorporating multiple servers and a range of server allocation strategies. Based on the analytical models, we will propose server allocation and peer-to-peer chunk-exchange strategies that improve streaming quality (e.g., we will study how to utilize multiple servers to mitigate the performance degradation caused by peer churn)
StatusFinished
Effective start/end date8/16/125/31/14

Funding

  • National Science Foundation (NSF): $190,448.00

Fingerprint

Servers
Chucks
Analytical models
Degradation