| IP networks, Internet as one of their representations, are becoming basic facilities of our society. As the scale of IP network expands continuously, network bandwidth increases in double and its structure becomes more and more inhomogeneous and complex, it will be of great benefit for network performance management and network monitoring to detect network performance on IP network precisely and real-time. Performance measurement on IP network will provide scientific reference to IP network for effective network management and controlling, reasonable design and use, development of protocols, devices and applications.Based on several network test and measurement projects, this paper mainly discusses technologies of bandwidth and traffic measurement. Traffic models based on measurement on operated networks are proposed and performance measurement technologies applied to network monitoring system are also discussed. The main achievements during the period of pursuing my Ph.D. degree are described as follows:1. An efficient and precise methodology of bandwidth measurement is proposed for avoiding effect of cross traffic and error accumulation hop-by-hop.Existing tools for bandwidth measurement such as pathchar, clink, pchar, nettimer and bing have been downloaded and employed for measurement experiments at the operated network of ICT. The results of these experiments show that the cross traffic and error accumulation are the main causes for effecting the precision of measurement. An efficient and precise bandwidth measurement algorithm is proposed. It can avoid effect of cross traffic and error accumulation, reduce workload imported by measurement and improve measurement speed. This algorithm can get performance metrics such as the packet loss rate, round trip time, and latency variation according to the returned status of the measuring probe packets.2. Two traffic measurement system SNMPER and LANMon are designed and implemented.SNMPER, based on SNMP, is implemented according to framework of SNMPv3 and compatible to SNMPvl and SNMPv2. LANMon measures traffic metric by capturing packets. There are two measurement modes for LANMon, with model 1 precision of traffic metric can achieve millisecond level and with model 0 it can achieve traffic measurement in different protocol layers. LANMon adopts both file and database store methods. File stores the captured packets and database stores the index of the location of captured packet in file according to the protocol layer. Different with the general protocol analyzer, LANMon is suitable to long term performance measurement on large-scale networks. SNMPER and LANMon are applied to the actual network measurementn and provid experiment data for long term traffic characteristics analysis.3. A periodic logarithm traffic model is proposed which is suitable to apply into performance analysis of real time applications.A periodic logarithm traffic model is proposed based on massive experimental traffic data from measurement on operated networks. No restrictive assumption ismade on traffic in this traffic model and bursting characteristics is described in it. An expression of R(t) = aln(0t+l) describes the model, where a denotes burstingrange and ft denotes bursting traffic appearing time. Transporting capacity ofinputing and outputing interface, buffer size, processing ability of switch unit of router are estimated under the traffic model. It also predicts the maximum and minimum end-to-end delay in the traffic model.4. A traffic forecasting model is proposed.Based on LAN/WAN traffic measurements, a traffic forecasting model is proposed. Traffic is separated into two parts. One is a periodic function of time and the other which obeys normal distribution is irrelevant to time. Based on the above decomposition, a traffic forecasting algorithm is designed which can forecast the dynamic traffic at any time.5. An efficient SNMP polling policy is proposed and a hierarchical framework of network monitoring system based on network perfor...
|
PDF Full Text Request