Experimental comparison of neighborhood filtering strategies in unstructured P2P-TV systems

Stefano Traverso, Luca Abeni, Robert Birke, Csaba Király, Emilio Leonardi, Renato Lo Cigno, Marco MelliaP2P 2012 · OpenReview

Abstract

P2P-TV systems performance are driven by the overlay topology that peers form. Several proposals have been made in the past to optimize it, yet little experimental studies have corroborated results. The aim of this work is to provide a comprehensive experimental comparison of different strategies for the construction and maintenance of the overlay topology in P2P-TV systems. To this goal, we have implemented different fully-distributed strategies in a P2P-TV application, called Peer-Streamer, that we use to run extensive experimental campaigns in a completely controlled set-up which involves thousands of peers, spanning very different networking scenarios. Results show that the topological properties of the overlay have a deep impact on both user quality of experience and network load. Strategies based solely on random peer selection are greatly outperformed by smart, yet simple strategies that can be implemented with negligible overhead. Even with different and complex scenarios, the neighborhood filtering strategy we devised as most performing guarantees to deliver almost all chunks to all peers with a play-out delay as low as only 6s even with system loads close to 1.0. Results are confirmed by running experiments on PlanetLab. PeerStreamer is open-source to make results reproducible and allow further research by the community.

Introduction: The Challenge of Live P2P-TV

Live video streaming over Peer-to-Peer (P2P) networks is a complex balancing act. Unlike file sharing, P2P-TV requires video "chunks" to arrive before strict playback deadlines. In "mesh-based" systems, peers form an overlay topology to trade these chunks. The performance of these systems is fundamentally dictated by this overlay topology—specifically, who is connected to whom. While many theoretical models have proposed ways to optimize these connections, empirical validation in large-scale, controlled environments has been scarce. This work addresses this gap by investigating how different "neighborhood filtering" strategies—rules for adding or dropping neighbors—affect user Quality of Experience (QoE) and physical network load.

The PeerStreamer Platform

The researchers developed and utilized PeerStreamer, an open-source P2P-TV client, to test various strategies. PeerStreamer's architecture allows for modular implementation of chunk trading protocols and overlay management, as shown in the diagram below.

PeerStreamer Architecture

The study evaluated two main functions:

  1. Add Filters (Neighbor Selection): Criteria included random selection (RND), weighting by upload capacity (BW), proximity via Round Trip Time (RTT), and chunk offering rate (OFF).
  2. Drop Filters (Neighbor Removal): Criteria included random dropping (RND), proximity-based dropping (RTT), and performance-based dropping (RXC, based on chunks successfully received).

A Blacklisting policy was also implemented to temporarily ban peers showing high packet loss or late deliveries.

Experimental Methodology

The study utilized two primary environments for evaluation:

  • Controlled Lab Testbed: A setup at Politecnico di Torino involving 204 PCs running 1,020 simultaneous peer instances. This allowed for precise control over network variables like latency, bandwidth, and packet loss across four simulated "subnets".
  • PlanetLab: A global research network where they deployed 449 nodes worldwide to validate their findings under real-world, unpredictable Internet conditions.
  • Video Source: The "Pink" Aerosmith video sequence, encoded with H.264/AVC at 1.2 Mb/s, was used as the benchmark stream.

Key Insights: Beyond Random Selection

  • Limitations of Random Selection: While robust, random peer selection performs poorly when neighborhood sizes are small. It fails to effectively "find" and utilize high-capacity peers, leading to significant frame loss.
  • The "RTT-RXC" Synergy: The most effective strategy combined proximity-based adding (RTT) with performance-based dropping (RXC). This localizes traffic to reduce network costs while ensuring that only contributing peers remain in the neighborhood.
  • Essential Blacklisting: In "lossy" scenarios where network paths are unreliable, blacklisting underperforming peers was the "silver bullet" that allowed almost all peers to receive a perfect stream, even at 90% system capacity.
  • Risks of Over-Localization: Using RTT for both adding and dropping peers can be too aggressive. It can lead to "poors with poors" clusters where low-bandwidth peers are stuck together, unable to sustain the stream.

Performance Results

The RTT-RXC strategy, enhanced with blacklisting, achieved an average SSIM (Structural Similarity Index) > 0.985 at a system load of 0.9. The system maintained a remarkably low play-out delay of just 6 seconds, which is exceptional for an unstructured mesh P2P system.

Furthermore, RTT-aware policies significantly reduced the "network cost" (the physical distance data traveled) without sacrificing video quality. As shown in the CDF below, strategies incorporating RTT for neighbor selection drastically shift the network cost distribution towards lower values compared to random or bandwidth-based strategies.

CDF of Network Cost for different filtering strategies

Conclusion

This experimental study demonstrates that simple, distributed neighborhood filtering strategies can drastically improve P2P-TV performance. By combining network distance with peer contribution metrics, systems can achieve high quality and low latency with minimal overhead. Future work will explore incentive-compatible strategies to prevent free-riding and integrate advanced security measures against malicious nodes.

Summary updated

Abstract

P2P-TV systems performance are driven by the overlay topology that peers form. Several proposals have been made in the past to optimize it, yet little experimental studies have corroborated results. The aim of this work is to provide a comprehensive experimental comparison of different strategies for the construction and maintenance of the overlay topology in P2P-TV systems. To this goal, we have implemented different fully-distributed strategies in a P2P-TV application, called Peer-Streamer, that we use to run extensive experimental campaigns in a completely controlled set-up which involves thousands of peers, spanning very different networking scenarios. Results show that the topological properties of the overlay have a deep impact on both user quality of experience and network load. Strategies based solely on random peer selection are greatly outperformed by smart, yet simple strategies that can be implemented with negligible overhead. Even with different and complex scenarios, the neighborhood filtering strategy we devised as most performing guarantees to deliver almost all chunks to all peers with a play-out delay as low as only 6s even with system loads close to 1.0. Results are confirmed by running experiments on PlanetLab. PeerStreamer is open-source to make results reproducible and allow further research by the community.
BibTeX
@inproceedings{traverso2012experimental,
  author = {Stefano Traverso and Luca Abeni and Robert Birke and Csaba Király and Emilio Leonardi and Renato Lo Cigno and Marco Mellia},
  title = {{Experimental comparison of neighborhood filtering strategies in unstructured P2P-TV systems}},
  booktitle = {P2P},
  year = {2012},
  url = {https://openreview.net/forum?id=dnvZbWYNjZ},
}
Figure