| Along with the rapid development of computer networks, the performance of networks is facing more and more challenges and concerns. It’s essentially necessary to pay more attentions on how to infer the network performance efficiently and improve the performance. The current researches on the network performance inference mainly focus on network delay, loss rate and other performance metrics and have already made achievements. But there are still some important problems waiting to be solved, including too large probing cost and bad inflection to time changes. These problems all make a bad influence for infer the network performance rapidly and correctly.In this paper, based upon the simulation and analysis of current network performance inference algorithms, we propose a mathematical-programming based algorithm respectively for the performance inference of small networks and large networks according to the characteristics of networks with different scales. With the advantages of mathematical programming, the problems in network performance inference are solved easily. Our algorithm is designed specially for inference of link loss rate. After getting the information of network topology and dynamic probing, the loss rate inference problem is modeled as a mathematical programming problem. Therefore we can get the inference result by solving the mathematical programming problem with the advantages such as lower probing cost, more accurate result and shorter time consuming. We also carry out experiments to examine the correctness of this network performance inference algorithm and analyze the performance of our algorithm comparing to current ones and real network. |
