Predicting QoS Characteristics on Wireless Networks

Nowadays, wireless network traffic exhibits peculiar and unpredictable behaviors that can be attributed to the combination of different factors: the varying number of users, the capacity of the network access points, network traffic matrices, users´ mobility, background traffic, diversity of applications, services and user behaviors. This scenario creates an increasing need for tools that can cope with the continuous appearance of new applications and QoS requirements and is able to aid in the global management tasks of wireless IP networks. This paper proposes a novel modeling approach for wireless networks that is able to predict network quality of service (QoS) parameters, based only on measured or synthetically generated traffic, without requiring any information on the network structure or on the mobility patterns of network users. The solution to a specific modeling problem, using the proposed framework, is a two-step procedure: in the first step, the network model is constructed from a set of (past) values of the inbound/outbound traffic at the various network access points and a set of values of corresponding network QoS metrics; in the second step, the previously constructed model is used to predict (future) values of the QoS metrics based only on new inbound/outbound traffic values. This framework does not impose any restrictions to the type of metrics that are selected to characterize QoS. The results obtained show that this approach is able to achieve excellent performance for QoS prediction at multiple wireless access points, even when there are significant changes in the number of network users.