Kernel methods aplication to compare diferent rate distribution in Generalized Markov Fluids

BAVIO, JOSÉ; MARRÓN, BEATRIZ

Bahía Blanca

Congreso; XV Congreso Dr. Antonio Monteiro; 2019

Departamento de Matemática- Universidad Nacional del Sur

Computer networks as internet are widespread around almost every technological development. Networks architectures and designsrequires good estimation of traffic moving along the network in time,in order to share and assign resources in an optimal way.Traffic sources can be modeled with Generalized Markov Fluids(GMF), which is very versatile and has good properties. With GMF,channel occupation or resource requires can be estimated by the calculation of Effective Bandwidth (EB).Previous work on EB estimation using traffic traces, required adeep knowledge about modulating chain and strong assumptions onthe support of some relevant distributions within GMF definition. Thisrestrictions can be solved by applying kernel methods and reconstruction of the modulating chain is possible with very little prior knowledgeand broad assumptions.In this work, we compare EB estimations using kernel methodswhen the distribution associated to each state of chain is both, exponential and normal, and we also compare the goodness of infinitesimalgenerator reconstruction in each case.