Modeling movements of users is important for simulating wireless networks, but current models often do not reflect real movements. Using real mobility traces, we can build a mobility model that reflects reality. In building a mobility model, it is important to note that while the number of handheld wireless devices is constantly increasing, laptops are still the majority in most cases. As a laptop is often disconnected from the network while a user is moving, it is not feasible to extract the exact path of the user from network messages. Thus, instead of modeling individual user's movements, we model movements in terms of the influx and outflux of users between access points (APs). We first counted the hourly visits to APs in the syslog messages recorded at APs. We found that the number of hourly visits has a periodic repetition of 24 hours. Based on this observation, we aggregated multiple days into a single day by adding the number of visits of the same hour in different days. We then clusteredAPs based on the different peak hour of visits. We found that this approach of clustering is effective; we ended up with four distinct clusters and a cluster of stable APs. We then computed the average arrival rate and the distribution of the daily arrivals for each cluster. Using a standard method (such as thinning) for generating non-homogeneous Poisson processes, synthetic traces can be generated from our model.
展开▼
