ENGLISH ABSTRACT: In this thesis, we consider the multi-constraints QoS routing problem in IP networks. Namely, we
consider the problem of minimizing the path delays on IP networks. We use genetic algorithms to
perform the optimization, some penalty function methods and the simulated annealing method for
handling the problems constraints.
Our aim is to compare the performance of di erent penalty function methods and the simulated annealing
method. The penalty function methods under consideration include penalty methods with
non-stationary as well as stationary penalty coe cients. The basis for doing the comparisons are the
maximum link and path delays, the maximum and average path length, and the CPU time.
We used four virtual networks as test examples. We found that, generally, the performances of the
simulated annealing method, the dynamic and co-evolutionary penalty function methods are better
than the performances of the adaptive, annealing and the static penalty function methods. Dynamic
coe cients seem to have a slight edge over stationary coe cients. Simulated annealing turned out to
be the slowest of the approaches investigated.
AFRIKAANSE OPSOMMING: Hierdie tesis ondersoek hoe om die multi-beperking QoS (\quality of service") roeteringsprobleem vir
IP netwerke op te los. Meer spesi ek, die doel is om die netwerkpadvertragings te minimeer. Genetiese
algoritmes word gebruik om die probleem deur middel van optimering op te los, en die multi-beperkings
word hanteer met behulp van boetefunksies. Daar word ook gekyk na die tempersimulasie benadering
(\simulated annealing").
Die doel van die tesis is om die boetefunksies en tempersimulasie te vergelyk. Beide konstante en
nie-konstante boetefunksies word ondersoek en nuwe konstante boetefunksies word geformuleer deur
die nie-konstante boeteko e si ente vas te pen. Al hierdie metodes word gemeet deur te kyk na die
maksimum skakel- en padvertraging, die maksimum en gemiddelde padlengte, en die verwerkingstyd.
Vier virtuele netwerke word gebruik as 'n toetsraamwerk. Die uiteindelike gevolgtrekking is dat die
verskillende boetefunksies rofweg dieselfde antwoorde produseer. Nie-konstante ko e si ente presteer
ietwat beter as konstante ko e si ente. Die tempersimulasie was aan die einde van die dag, die stadigste
benadering waarna gekyk is.
