Research Of Content-based Flash Data Dissemination Technology

Data dissemination is a key technology for resource aggregation and sharing in the distributed environment. Content-based flash data dissemination aims to transmit medium-and small-sized data from one or more source nodes to several target nodes in a short time, according to the interests and requirements of the target nodes. Overlay mechanism provides an efficient infrastructure for content-based dissemination: the interests of the target nodes are expressed through the overlay networks; then data produced by source nodes is matched and filtered by the overlay and forwarded to the corresponding target nodes.Topology construction and maintenance, routing algorithm as well as matching algorithm design are fundamental problems of content-based flash data dissemination. Existed structured overlay topologies, though having efficient routing rules, exhibit weak scalability and flexibility; on the other hand, unstructured overlay topologies, owning simple and flexible topology construction process, however, cannot meet the timeliness requirements of flash data dissemination. According to the above limits of existing methods, three coherent sub-problems of the content-based flash data dissemination are deeply studied: scalable topology construction and maintenance methods, loop-free message routing algorithms, as well as the data attribute matching methods.Firstly, considering the efficient topology construction problem of content-based flash data dissemination in a dynamic environment, a content-based double layered overlay (CBDLO) is proposed. CBDLO is a compound topology composed of a lower unstructured topology and many upper attribute topologies, each of which is a distributed balanced binary tree (DAVL) that is made up of virtual nodes. In the CBDLO topology, each value space of the subscription attribute is divided into many joined but not intersected subspaces, and each subspace is designated to a virtual node, which is mapped into a real node of the network. The virtual nodes finally consist of a distributed AVL tree. Detailed simulations are carried out on the Peersim simulator based on the event driven model. Simulation results show that CBDLO can well adapt to the dynamics of the network, and can scale gracefully as increasing of the size of the network.Secondly, to reduce the routing message cost on the lower unstructured topology, a new content-based routing algorithm with labeled trace (CRAWL) is proposed. CRAWL is designed to transmit data in a random walk manner, so as to locate the matched upper attribute topology. During the random walk process, the recently traveled node is added to the routing message as a trace so as to avoid the routing loops. Meanwhile, a path restoring mechanism is also proposed to avoid the routing deadlock posed by trace. Simulations indicate that CRAWL algorithm significantly reduces the cost of the message and thus improves the performance of dissemination.Thirdly, in order to increase the efficiency of accurate matching process in the distributed AVL, a counter-based distributed matching algorithm CDM is proposed. Each attribute in the data corresponds to a distinct distributed AVL tree; the same data is matched in different distributed AVL trees in parallel. Finally, the data will be forwarded to all the matched target nodes. Each target maintains a counter for each data to confirm that the data matches all requirements of its subscription's attributes. Simulations show that CDM algorithm greatly reduces the matching cost, and is an excellent candidate for flash data dissemination.