Research On Network Coordinate System Based On Large Scale Distributed Networks

In the past few years, with fast development of the network technology, the great potential of the internet infrastructure was further developed. Global distributed networks and applications have been formed. Large-scale distributed network research has become one research hotspot of the current internet. Many distributed applications such as C DN(Content Distributed Networks) and the bitstream, in the process of providing services, need to know the network distance of the node to node in order to improve the service quality. At the same time, the current internet has developed into the cloud computing era. A cloud environment often faces the basic problem on how to choose the server. How to select the appropriate server node for the terminal node is which a urgent problem to be solved for cloud service providers.Network coordinate system is a new kind of internet distance prediction scheme. It has characteristics of high scalable and low overhead. Compared with traditional direct measurement method, measurement of time complexity can be reduced from O(N2) to O(N). Based on this point and the deficiency of the existing network coordinate system in the accuracy and scalability aspects, this paper aims to develop a new network coordinate system and apply the technology of the network coordinate system to the server selection problem in cloud environment.At first, we put forward a novel distributed coordinate system algorithm DISCS based on N RPC A(Nonnegative Robust Principal Component Analysis). DISCS eliminates the influence of measurement errors and abnormal points through extracting the main component from the region dataset. Then through mult iple iterations, DISCS obtains the accurate node coordinates. Through these coordinates, DISCS can calculate the distance between the node to node.Based on the idea of DISCS algorithm, we put forward a new distributed network coordinate system algorithm which called WDCS. WDCS combines the idea of principal component analysis and thoughts of the physical mass-spring system. First, WDCS extracts the principal component from the region distance matrix of nodes. Then, it uses the idea of physical mass-spring system to update coordinates. Through edge weights of the network, each reference node is given different confidence to eliminate errors of coordinates. In terms of accuracy, DISCS is better than WDCS. But WDCS saves a large amount of computing time than DISCS.Based on WDC S, we put forward the idea of multi- neighbors convergence. In the update process of nodes' coordinates, WDCS changes reference nodes every fixed time slices. We call this algorithm WDCS- MNC which combined with principal component analys is, multi- neighbors convergence and physical particle spring system.We then expound the network coordinate system application in the server selection problem within a cloud environment. This paper proposes a new cloud server selection strategy which is called DEAL. DEAL is a kind of distributed server node selection strategy, which has high scalability. By setting the intra-domain transit traffic penalty coefficient, DEAL takes account of the interests of internet service providers to the algorithm. DEAL allows internet service providers to choose flexible mapping strategy from terminal nodes to server nodes.In this study, the proposed method is compared with existing latest methods through simulation experiments. Several methods are proposed in this paper. In different network environments, their performance is different. In all aspects of the evaluation, each one has its advantages. But they are all better than existing methods. DISCS, WDCS and WDC S-MNC can accurately compute node coordinates, and estimate the distance between nodes. DEAL can quickly and accurately complete the mapping strategy from end users to server nodes.