Research On Topology Optimization Of Gossip-based Self-organization Overlay Multicast

Large-scale multicast data distribution under military circumstances plays an important role in future information warfare. Modern military communication is featured with large scale, drastic dynamics, heterogeneity and high QoS demand. In particular, the combat environment brings about drastic dynamics in military network, which far overruns the limit of conventional network management. Employing overlay multicast data distribution on application layer can better complete the large-scale data distribuition task in military evironment. By employing the overlay networks built on top of physical network, we can get more support in aspects of robustness and flexibility, thus providing the communication networks with better adaptability in complex and dynamic environment.In contrast, some natural networks like the biological network, though giant in size, have strong adaptation to cope with various emergencies. Introducing the adaptation of natural network into military communication network can equip the latter with better adaptation. The thesis applies gossip mechanism developed from theory of epidemic spread as the major approach to construct and manage overlay topology, and optimizes the multicast overlay topology. The thesis takes the combination of unstructured overlay network and network coding as the main method to realize overlay multicast data distribution in military environment, elaborates on the topology optimization involved in this method, studies several unsolved yet urgent key issues in military application and proposes effective solutions.(1) Study of the quick overlay topology convergent mechanism based on self-adaptive gossipGossip-based unstructured overlay topology can adjust itself with the network environment change, enabling the overlay topology to converge to different structures. The topology convergent speed reflects the adaptability of overlay network in dynamic environment. The faster the convergence is, the better the adaptablility is. This paper analyzed the node behavior characteristics in the convergence process and found the flaw of fixed cycle gossip. Then the paper proposed an adaptive cycle mechanism to control the exchange of gossip packets. This mechanism can improve the data exchange efficiency and save the network resource so that some nodes can accelerate the data exchange locally and therefore the convergence speed of overlay topology is increased.(2) Study of connectedness in self-organizing overlay network evolvingGuaranteeing the connectedness of network is an important task in overlaytopology management. The unstructured overlay topology management changes the nodes neighbor lists to match the change of underlying physical network. This paper analyzed the overlay topology evolving process in overlay topology management methods based on gossip mechanism and found the local-area characteristics of actual network and neighbor selection strategy of nodes in unstructured topology may break network connectedness of overlay topology. After analyzing the key factor of breaking connectedness, the paper proposed a simple yet convenient links replacement algorithm based on local connected domain optimization. Furthermore, by combining different simplex topology management mechanisms, the paper presented Multiple Lists Membership Management (MLMM) and gave an example based on MLMM, C-MLMM, which showed the efficiency of MLMM in assuring topology convergence speed and network connectedness as well.(3) Study of distributed network coding construction based on gossipPrevious centralized approaches on network coding construction need global topology knowledge and are difficult to be practically used in large-scale or dynamic network. And the randomized network coding approach is a practical distributed network coding approach, but it employs all nodes to code and consumes enormous computing power. However, the data distribution in military communication network demands a more effective distributed network coding construction approach. Hence this paper decomposed the network coding solution into several basic "request-respond" processes, summarized the typical behavior of different kinds of nodes in the network coding scheme. Then applying gossip mechanism that can obtain the adjacent node information quickly, the paper proposed a distributed network coding construction approach with much fewer coding nodes. Simulation showed that this approach has excellent performance, which achieves the approximate throughput compared with distributed Randomized Network Coding, but only employing about 10% of nodes to code.(4) Study on dynamic selection and maintenance of coding supernodeIn the network coding multicast solution, coding nodes need consume extra computing power. Nodes with stronger abilities may be chosen as candidates for coding, called Supernodes. The prior usage of supernodes in network coding can avoid overload or breakdown caused by the inability of coding nodes. This paper analyzed the characteristics of the supernodes for network coding and presented a dynamic supernode selection and maintenance policy applying gossip mechanism. The policy is proved to suit for network coding multicast.(5) Study of resource sharing and competition among multiple sessions in network coding multicastThere are different network services in military communication application and the problem of resource sharing and competition is inevasible. This paper analyzed the specific characteristics of resource sharing and competition in network coding multicast. To share resources among multiple sessions efficiently, a general integrated membership management approach was presented. Then games theory was used to study resource competition behavior in different sharing models and the paper found the necessity of introducing certain trust mechanism to improve resource sharing.In a word, the thesis studied large-scale data distribution in the complex and highly dynamic military environment, proposed a data distribution multicast scheme combing gossip-based unstructured overlay topology and network coding suitable for military application, and thus laid sound foundation for the capability improvement on data acquisition and distribution in military communication network.