Method of estimating of end-to-end performance distribution using conventional managed objects

Demands for high-quality inexpensive IP networks are exploding these days. In order to provide satisfaction (i.e., a cost-effective network) to users, network managers should know the end-to-end performance. Moreover some applications permit occasional degradation of performance if it makes the service cheaper Thus, it is important to know the probability distribution of the end-to-end performance, e.g., the 10%ile. The methods of determining the end-to-end performance are classified into three groups: 'Measuring methods', 'Calculating methods', and 'Combined methods'. This paper proposes a new 'combined method' of estimating the probability distribution of the end-to-end performance using (i) managed objects (MOs) collected in conventional IP network management and (ii) the mapping between the MOs and the end-to-end performance obtained by analyzing their values measured beforehand This method can estimate the probability distribution of the end-to-end performance, so a flexible threshold can be made. In daily management, this method only needs to collect conventional MOs, so it does not generate extra traffic or affect the other traffic. Variable selection is introduced into regression analysis and it can find a bottleneck in the measurement path