Admission Control Of P2P-based Media Delivery Network

With wider application of P2P-based Media Delivery Network (P2P-based MDN), much work has been done to improve the system'performance and efficiency in the process of media contents sharing. Our work is just developed based on this background. We study the admission steps, admission strategy and service strategies in P2P-MDN. Admission policy, as the entrance for the VOD service, has great impact on the workload of the whole system and quality of service (QoS). When admission policy is fixed, the service strategies for service peers will determine the simulation parameters of the system directly and also influence the performance of the system in every aspect.Our major work includes:1. We propose a specific partially observable Markov Decision Process (POMDP) model for admission control of P2P-based MDN. Based on this model, we provide observation-based randomized parameterized policy to conduct admission control. Observation-based policy can promote the system's performance in real time aspect and parameterized policy is designed for the optimization. We optimize our policy by applying the policy-gradient algorithm. Based on partial information, policy optimization can increase the accuracy of controller's judgments and increase the average payoffs of the providers in long-term operation.2. Novel source selection policy and bandwidth allocation policy are designed to reduce service delay and provide high-quality service. A statistic-based selection (SBS) policy is proposed based on our model and simulation method. In SBS, the overload probabilities of PNs (Peer Node) are estimated based on the simulation statistical data, and the original source selection strategy is ameliorated accordingly. Bandwidth allocation method is designed to avoid the problems existing in traditional methods, in which every peer allocates its bandwidth for every other peer. We propose three different allocation strategies. In partial service policy (PSP), every PN allocates the bandwidth to make the service rates directly proportional to the requests'arrival rates.3. Based on the theoretical results, we design experiments to compare our optimal policy obtained through simulation with the Best Effort Service strategy. It can be seen that, optimal policy reduces unnecessary resource waste and saves the cost in operating the system for media service providers. Meanwhile, indirect rejection rate sharply reduces with a higher success rate in accepted services. The system can also keep a comparatively high bandwidth utilization rate, higher than 50%.4. The quality and efficiency of the system can be further improved by meliorating the service scenario of the PNs. We design four different scenarios to compare. The comparison results indicate that: (1) service scenario which allocates bandwidths according to the demands can utilize the system's bandwidth resource in a better way and allow more users to enjoy services in the system. This method brings more economic benefits for the service providers; however, the proportion of accepted services being rejected indirectly by other remote PNs also goes up under this method. (2) SBS can not only reduce the cost in system operation and raise the utilization of the system resource, but also adjust load balancing of the system, so the indirect rejection rate reduces under this strategy. To sum up, the service scenario with both PSP and SBS is the best combination in improving the performance of the system.