Research On Network Flows Measurement Algorithms Based On Sampling And Hash Techniques

The Internet is a global network connected by millions of computers, with therapid development of network technology, the network performance are enhancingunceasingly, users and network application is increasing gradually, network flows arebecoming more and more large, the network behavior is also becoming more and morecomplicated. Mastering the internet behavior is the important premises for manyresearch works such as network planning, network management, network security, newnetwork protocol and network application design, etc.The measurement of network flows is the most effective way to understand thenetwork behavior, is the foundation and prelude to control the internet, also the basis toredesign and upgrade the existing internet. Because of the rapid development ofhigh-speed net and huge quantity of network flows,it is very difficult to measure flowreal-timed and online. So a variety of flow measurement techniques and algorithmsbecome a research hot-spot.In this paper, we survey the achievements obtained in the fields of flow algorithmsby using sampling technique, and put forward improved stratified sampling technique.After, we change the traditional stratified strategy by reconfiguring parameters andexploring theory, we use simple linear estimation method to inferring originalinformation. Furthermore, by using Φ discrepancy examine method, we carry on theaccuracy performance comparison between improved stratified sampling technique andother sampling methods form packet size distribution aspect.Later on, we analyze the current situation of flow measurement algorithms researchwhich based on hash technique, then according to the characteristic of TCP flow, a newpacket filtering technique CTBF(Counting Time-out Bloom Filter) is proposed togenerate the statistics of the distribution of TCP flow length. This method can improvethe performance of network and the efficiency of the resource usage of measurementsystem by handling fmeasurement lows with different strategy according to its differentfeature.At last, after considering the advantage and disadvantage of sampling technique and hash technique, we propose a measurement method that combines both techniques.Furthermore, according to the characteristics that the network flow statistics show astrong heavy-tail distribution, we give FS-CBF(Flow Sample-Counting Bloom Filter)method to realize elephant flows identification, which is combined period sampling andCBF technique.