Network planning guaranteeing end-to-end overload probability for stochastic traffic demands

Planning a communication network is a very challenging task, because network traffic is not constant but fluctuates heavily. Overestimating the traffic volume leads to an expensive solution, while underestimating it results in a poor Quality of Service (QoS). In this paper, we propose a new approach to solve the network planning problem under stochastic traffic demands, which guarantees the overload probability of an end-to-end traffic demand to be bounded by a pre-determined value. The problem was first formulated as a chance-constrained programming problem, in which the capacity constraints need to be satisfied in probabilistic sense. We then propose two heuristic algorithms, which 1) determines the overload probability on each link so that the end-to-end overload probability of a traffic demanded is guaranteed and 2) solves the routing and capacity allocation problem for given stochastic traffic demands. The experiment results show that the proposed approach can significantly reduce the network costs compared to the peak-load-based approach, while still maintaining the robustness of the solution. This approach can be applied to networks carrying different flows with different QoS requirements.

Subjects

chance constrained programming

genetic algorithm

Network planning

stochastic traffic demands

Options

Network planning guaranteeing end-to-end overload probability for stochastic traffic demands