Research On Dynamic Content Management Based On P2P Network And Cloud Computing

With the improvement of network, hardware and software, some terminals are more and more powerful to participate in the network construction. These terminals can share the work of servers, improve the scalability of the network, reduce the cost of service providers, and promote the development of the distubuted network, such as P2P network. Furthermore, more and more new terminals with limited resources, such as mobile phone and PDA, has to rely on the server, which promotes the appearance of cloud computing. But these two kinds of terminals are not separate. For the heterogeneity of the network, we can give different terminals different functions based on their abilities, so they can play different roles in the network. It is one of the new developing directions of future network.But it will bring new problems when user's terminals participate in network construction and maintaining. Because there are amount of peers and resources in the network, the key problem of dynamic content management is to build logical topology and relevant protocols, so that we can locate resources and deliver updates effectively. Real-time of operation, robustness and load-balance of system are basic demands of distributed dynamic content management network, but the heterogeneity and unreliability of peers will become a balk to above demands.To construct a network for dynamic content management meeting all above demands, we focus on network topology design and resource discovery and update delivery protocol design. In order to improve the robustness and load balance of the system, we study real-time and consistency as well as the effects upon network topology and performance introduced by dynamic changes of participant peers'demands.Main contributions in my dissertation are listed as follows.1. By using massively multiplayer online game (MMOG) as a typical application of dynamic content management, we propose a communication platform based on P2P for MMOG.A model with good performance of load balance was proposed which is called Slide. In Slide, normal peers help each other for resource discovery, while super peers assist normal peers only when there was no other normal peer. In this method, Slide can discover resources step by step, so the load of super peer can be decreased and redundant communication can be avoided. Normal peers deliver updates to all relevant peers directly, in order to further reduce the load of super peer and improve the capability of real-time. The consistency check method based on optimistic synchronization was proposed to ensure the consistency under low traffic and improve the robustness of the system. Theoretical analysis and simulation results showed that the request of real time is met in Slide, while the consistency and load balance can be achieve too. We also used Slide as the basic communication platform to build a simple MMOG.2. Dynamic content management in P2P network will be discussed in the follow three fields in depth.In order to provide a stable network for dynamic content management, a double-layer structure was designed to combine resource discovery with update delivery, which constructs the Stable Content Distribution Network (SCDN). Resources are discovered based on demand, so the performance is improved and the overhead is decreased. The hybrid Push/Pull method was proposed for update delivery, in order to ensure the consistency of content and replicas. The protocol of peer management was proposed for robustness too. By using Markov Chain, we analyzed the performance of SCDN, and the further simulation results showed that SCDN can work real time and ensure the consistency and robustness at the same time.Based on SCDN, a content-peer double-cluster model was proposed using the limitation of peers'demands. By constructing the list of relevant peers to form a cluster, two-step search method was proposed to improve the performance of resource discovery, especially the search for relevant content. Simulation results proved the performance is better in our model, and the traffic and storage space are also smaller.In order to further explain why our models can improve the performance of resource discovery significantly, the small-world theory were brought. The effects of different long-distance connections were analyzed. Results showed long-distance connection based on history has good performance of resource discovery in real network, and the overhead is also low. Though long-distance connection based on Kleinberg theory has the best performance in theory, it is not the best in real network. 3. In order to adapt to the diversity of network terminal and the appearance of cloud computing, we combined P2P technology with cloud computing, and proposed a P2P-assisted cloud with low cost which is called PA-Cloud.Using P2P technology, the peripheral cloud is built around the core cloud to manage the dynamic contents, so the load of core cloud is decreased. As a result, the core cloud can focus on providing stable storage and backup. PA-Cloud can achieve the high efficiency in dynamic content management with significant low cost while ensuring the scalability and robustness of the system. PA-Cloud proposed the replica placement based on demand, the update method with low cost, and the adaptive mechanism for replica peer management. Simulation results demonstrated the effectiveness of PA-Cloud in comparison with pure cloud. With only 60% of servers in pure cloud, PA-Cloud can deal with the same workload while maintaining the quality of service. A good trade-off between cost and performance was achieved.