By Sudhir K. Routray, Instituto de Telecomunicações, Portugal
Optical transport networks are the backbone of the modern communication systems. During the planning phase, all the information of the network is not available. So, it is very difficult for the network designer to estimate the capital expenditure (CAPEX) needed for the network. Under such incomplete information scenarios statistical models are used for the estimations. There are several statistical models proposed by several researchers. However, there was no model for the link lengths.
The lack of a model for the link lengths was a key problem in the dimensioning of the optical transport networks. There are several link-dependent parameters and without the knowledge of the link lengths those parameters cannot be estimated with accuracy. We measured the link lengths of forty real optical transport networks and analyzed them for an appropriate statistical model. Initially, we thought of some common statistical distributions such as Poisson or Normal would provide the required model. However, we were surprised that the 'goodness-to-fit' results for these distributions were quite bad, and for several networks the results were not acceptable.
We wanted to test as many distributions as possible for an appropriate model. The search for a software which can provide the best match brought us to EasyFit. It was wonderful that EasyFit had 65 different distributions for continuous random variable, and it was tailor made to analyze the statistical distributions which we were looking for. We tested 61 relevant distributions for our model. We found that the Generalized Extreme Value (GEV) distribution was the overall best. Our work became quite straight-forward with EasyFit. EasyFit provides the parameters of the distributions along with the 'goodness-to-fit' results. We analyzed the parameters of GEV distribution using the outcomes of EasyFit. We were able to estimate the parameters of GEV distribution model from the basic information of the optical transport networks. Our model for link lengths in optical transport networks was accepted by the experts of the field. We want to extend this model for other networks.