An Adaptive Timeout Strategy for Profiling UDP Flows

With the increase of network bandwidth, more and more new applications such as audio, video and online games have become the main body in network traffic. Based on real time considerations, these new applications mostly use UDP as transport layer protocol, which directly increase UDP traffic. However, traditional studies believe that TCP dominates the Internet traffic and previous traffic measurements were generally based on it while UDP was ignored. In view of this, we mainly discuss the adaptive timeout strategy of UDP flows in this paper. Firstly, due to the significant differences in flow characteristics between the TCP flows and UDP flows, we expound and prove that the existing adaptive timeout strategies are not appropriate for UDP flows. Secondly, we present our adaptive strategy using Support Vector Machine techniques. We build six classifiers to accurately predict its corresponding maximum packet inter-arrival time and adapt its timeout value within the flow duration. Limited to its accurate rating, we present another concept of adjust accuracy rating which can probability-guaranteed(90%,95%,98%) to avoid long flow to be cut into short flows. The experiment result reveals that our adaptive strategy has the potential to achieve significant performance advantages over other widely used fixed and other adaptive timeout schemes.