| In recent years, as the scale of the IP network grows exponentially, the structure of network has become highly heterogeneous and complexity. IP network performance has become more and more complicated and difficulty to know. In addition, the continuous generation of the new information transmission technologies, the new network environment and the new network services, meanwhile users requirements of the network quality of service are also rising,In order to optimize network applications, improve network efficiency and provide a reliable theoretical basis, regardless of the network operators or Internet users, they need to know the overall network topology structure and performance parameters of network.In a real network, latency (ie: network distance) is a very important parameter and has become a fundamental property of network paths. Meanwhile it is closely relative with the network topology and routing. If obtained the time delay between the nodes, we would have improved the performance of network applications (such as media file sharing, content access network, etc.). Ping that can obtain the delay is the most , simple and intuitive method, but it is a low efficiency, large overhead, poor scalability mothod, and its time complexity is O(N2). Thus this paper presents the concept of IP network coordinate system, its basic idea is: let network distance space map into a geometric space, and the each network node has corresponding coordinate. Finally the distance between nodes can be predicted according to their spatial coordinates calculated from the formula. Therefore, the IP network coordinate system can greatly reduce the overhead of measurement.Currently, the literature proposed many different IP network coordinate algorithm. Even though these algorithms relative error is not large, but even small relative error, the performance of network applications will also have a very significant impact. Therefore, the focus of this paper is to improve the accuracy of the IP network coordinate system, including:1. The experience with selecting nodes that are in the different distance ranges as anchors shows that choosing the nodes which are close to the predicted network distance as anchors can improve the predicted network distance's accuracy. So, we propose a distance range-aware IP network coordinate system. The main idea is that according to the approximate range of the predicted network distance, we re-select the nodes that are in a similar range with the distance radius of the space as anchors and then, use the anchors to get its new range. In accordance with the process, the coordinate system constantly refines the network distance and dynamically adjusts the choice of anchors. The process dosen't stop until the prediction accuracy meet the requirement. Experiments show that this approach can not only effectively improve the prediction accuracy of distance, but to a certain extent, it can overcome inconsistencies of the prediction accuracy .2. In order to improve the prediction accuracy and reduce the computation overhead of R-Vivaldi, an enhanced distance range-aware IP network coordinate system based on -approximation(R -Vivaldi) is proposed. The main idea is that the distance range is computed with (triangle inequality). If the value of the distance that is computed by the R-Vivaldi's global coordinates is not in the range, it will be amend. Then it is put into R-Vivaldi to improve the prediction accuracy. Experiments show that this mechanism can not only effectively improve the prediction accuracy of R-Vivaldi′s distance but also reduce its computation overhead.3. The research found that TIV has become a major factor reducing the prediction accuracy. Thus in order to restain TIV's impact and improve the prediction accuracy, an fleetly regulative adjustment of the predict distance. The main idea is that base on the relationship between the landmarks and the node, then selecting the RR to amend the predict distance. |
PDF Full Text Request