Roteamento multicast multisessão: modelos e algoritmos
Multicast Technology has been studied over the last two decades and It has shown to be a good approach to save network resources. Many approaches have been considered to solve the multicast routing problem considering only one session and one source to attending session‘s demand, as well, multipl...
Na minha lista:
Autor principal: | |
---|---|
Outros Autores: | |
Formato: | doctoralThesis |
Idioma: | por |
Publicado em: |
Brasil
|
Assuntos: | |
Endereço do item: | https://repositorio.ufrn.br/jspui/handle/123456789/25734 |
Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
Resumo: | Multicast Technology has been studied over the last two decades and It has shown to be a
good approach to save network resources. Many approaches have been considered to solve
the multicast routing problem considering only one session and one source to attending
session‘s demand, as well, multiple sessions with more than one source per session. In
this thesis, the multicast routing problem is explored taking in consideration the models
and the algorithms designed to solve it when where multiple sessions and sources. Two
new models are proposed with different focuses. First, a mono-objective model optimizing
residual capacity, Z, of the network subject to a budget is designed and the objective is to
maximize Z. Second, a multi-objective model is designed with three objective functions:
cost, Z and hops counting. Both models consider multisession scenario with one source
per session. Besides, a third model is examined. This model was designed to optimize
Z in a scenario with multiple sessions with support to more than one source per session.
An experimental analysis was realized over the models considered. For each model, a set
of algorithms were designed. First, an ACO, a Genetic algorithm, a GRASP and an ILS
algorithm were designed to solve the mono-objective model – optimizing Z subject to a
budget. Second, a set of algorithm were designed to solve the multi-objective model. The
classical approaches were used: NSGA2, ssNSGA2, SMS-EMOA, GDE3 and MOEA/D.
In addition, a transgenetic algorithm was designed to solve the problem and it was compared
against the classical approaches. This algorithm considers the use of subpopulations
during the evolution. Each subpopulation is based on a solution construction operator
guided by one of the objective functions. Some solutions are considered as elite solutions
and they are considered to be improved by a transposon operator. Eight versions of the
transgenetic algorithm were evaluated. Third, an algorithm was designed to solve the
problem with multiple sessions and multiple sources per sessions. This algorithm is based
on Voronoi Diagrams and it is called MMVD. The algorithm designed were evaluated on
large experimental analysis. The sample generated by each algorithm on the instances
were evaluated based on non-parametric statistical tests. The analysis performed indicates
that ILS and Genetic algorithm have outperformed the ACO and GRASP. The comparison between ILS and Genetic has shown that ILS has better processing time performance.
In the multi-objective scenario, the version of Transgenetic called cross0 has
shown to be statistically better than the other algorithms in most of the instances based
on the hypervolume and addictive/multiplicative epsilon quality indicators. Finally, the
MMVD algorithm has shown to be better than the algorithm from literature based on the
experimental analysis performed for the model with multiple session and multiple sources
per session. |
---|