TKK | Tietoverkkolaboratorio | Julkaisut

Kaiyuan Wu: Load balancing of elastic data streams in cellular networks

Abstract of the Master's Thesis

The study describes a simple model of two adjacent cells, where the arriving rates are Poisson processes and the service time is exponentially distributed. Capacities of the base stations might be different, so both symmetric and asymmetric cases are considered in this thesis. Because of the 'elasticity' of data transfers in the data networks, we model the servers as processor sharing queues. The focus is to load balance the data flows generated by the users in the overlapping area of these two cells to minimize the average flow delay in the system. In [3], it has been proved that the optimal policy for two stations problem can be characterized by a switch-over curve, which strengthens our belief in finding such curves in our case. The static load balancing scheme is derived by the analytic method by queuing theory based on the model, in which two cells are modelled separately. The policy optimization is implemented by policy iteration, which is developed by Markov decision process. We repeat the iteration process until nothing changes, which indicates the appearance of the optimal policy. Later on, the results are illustrated with a few numerical experiments where different schemes are compared in terms of their mean delay performances.

The optimal policy can be characterized by a switch-over curve and it is even linear in the regarded problem. A rather easy-derived scheme is proposed because it can closely resemble the optimal policy, by which the mean flow delay in the system can be significantly reduced in theory.

Keywords:

Cellular networks, load balancing, elastic flow delay, MDP, processor sharing

Tietoverkkolaboratorio on nyt osa Tietoliikenne- ja tietoverkkotekniikan laitosta. Tällä sivulla oleva tieto voi olla vanhentunutta.