Abstract
This paper proposes MOPSO-DS, a novel data scheduling scheme for Peer-to-Peer (P2P) streaming systems. By utilizing multi-objective particle swarm optimization, the scheme addresses the inherent conflict between maximizing perceived video quality and optimizing network throughput. Unlike traditional single-objective heuristics, MOPSO-DS identifies a Pareto-optimal set of scheduling decisions that balance segment urgency with segment rarity. Experimental results from large-scale simulations demonstrate that MOPSO-DS significantly improves video continuity and Peak Signal-to-Noise Ratio (PSNR) while maintaining superior load balancing across the network.
Introduction: The P2P Streaming Challenge
In P2P streaming architectures, peers reduce the burden on central servers by sharing video data directly with one another. A critical component of these systems is the Data Scheduling module, which determines which video segments a peer should request from its neighbors and in what order.
As shown in the architecture above, the system must manage data flow between the source server, trackers, and a dynamic swarm of peers. The scheduling challenge arises from two conflicting objectives:
- Perceived Video Quality: To ensure smooth playback, the system must prioritize "urgent" segments—those nearing their playback deadlines.
- Network Throughput: To maintain a healthy swarm, the system must prioritize "rare" segments—those held by the fewest neighbors—to prevent them from disappearing and creating bandwidth bottlenecks.
Traditional algorithms often focus on one objective at the expense of the other, leading to either playback stutters or inefficient network utilization. MOPSO-DS treats this as a multi-objective optimization problem to find a more effective middle ground.
The MOPSO-DS Framework
Mathematical Modeling
The authors first formalize the scheduling problem by defining two fitness functions. $F_1$ focuses on maximizing the weighted sum of segments delivered on time, while $F_2$ aims to maximize the upload bandwidth utilization of all senders in the neighborhood. To calculate these values, the system tracks various parameters related to segment availability, deadlines, and peer bandwidth.
The table above details the symbols used to construct the optimization model, including variables for segment emergency ($P^j_{emergency}$), quality ($PSNR_j$), and neighbor availability ($a_{i,j}$).
Optimization Strategy
MOPSO-DS adapts the Particle Swarm Optimization (PSO) algorithm—typically used for continuous spaces—to the discrete task of data scheduling.
- Encoding: Each "particle" represents a potential schedule. The position vector is encoded such that each dimension corresponds to a video segment, and the value represents the specific neighbor selected to provide it.
- Pareto Optimality: Rather than seeking a single solution, the algorithm maintains an External Archive (EA) of "non-dominated" solutions. These represent the Pareto front, where one objective cannot be improved without degrading the other.
- Selection: At the end of each scheduling window, the system selects the candidate from the archive that best balances Quality of Experience (QoE) with high throughput.
Performance Evaluation
The researchers evaluated MOPSO-DS using an event-driven simulator with 2,000 peers and high-resolution video traces ("From Mars To China" and "Fugitive"). They compared the scheme against four classic algorithms: SSTF, WSS, LRF (Local Rarest First), and Min-cost.
Key Results
- Enhanced Video Quality: In the "Mars To China" trace, approximately 80% of peers using MOPSO-DS achieved a PSNR between 36 dB and 40 dB, significantly outperforming the 23–35 dB range of competing algorithms.
- Improved Continuity: MOPSO-DS maintained a Continuity Index (ratio of on-time segments) above 0.9 for the majority of peers, whereas other algorithms often fell to between 0.32 and 0.61.
- Superior Load Balancing: The scheme consistently achieved a lower Load Balancing Factor, indicating a more even distribution of work across the network, which prevents individual peers from becoming bottlenecks.
- Computational Efficiency: Despite its mathematical complexity, the execution time of MOPSO-DS remained low and comparable to simple heuristics, making it viable for real-time applications.
Conclusion and Future Work
MOPSO-DS demonstrates that treating data scheduling as a multi-objective trade-off problem yields superior results compared to traditional heuristics. By balancing urgency and rarity, the system achieves smoother playback and better network health. Future research will focus on adapting the scheme for Scalable Video Coding (SVC) and testing its robustness in environments with high peer "churn" (frequent joining and leaving).
Summary updated