Methodology for optimal link selection using multivariant normal distribution

Network throughput and bandwidth overprovisioning do not provide equal quality of service for all applications. Time characteristics of the network must also be taken in mind when deploying applications which are sensitive to delay and jitter, e.g. VoIP. By means of changing the path of traffic significant improvement of the time-variable can be achieved. In this paper we propose a methodology for optimal link selection based on modeling network characteristics using a Multivariate normal distribution model - cube model. This approach enables real-time identification of reliable links and provides input for a modified composite optimal path selection algorithm. Aside from this our model can be used for determining future network improvements to increase the quality of service.