Human movement patterns, mobility models and their impacts on wireless applications

Simulating human mobility is important in most areas that rely on human movements such as the performance study of mobile networks, disease outbreak control, city planning, etc. To test these applications, deploying a real testbed for a target scenario and running experiments would be the best solution but such experiments are extremely difficult. Thus to have a realistic mobility model and to run simulations with that model is invaluable. Numerous mobility models have been proposed until now but it is still questionable whether they are “realistic” enough to reflect human movement patterns. Since a mobility model is most useful when it is able to reproduce the mobility patterns realistically in any possible environments, defining “realism” and reproducing it is a key factor in mobility modeling. In this dissertation, we define the “real” human movement patterns as the heavy-tail flights and inter-contact times (ICT) which is defined to be the time duration until two mobile nodes meet again after meeting previously. These are important factors to determine the performance of human related applications since they govern the meeting probabilities among users. We have tested 8 well-known existing models and by varying the input parameters, we tried to fit their flight and ICT distributions to the distributions observed from empirical data sets. Unfortunately, none of them can match both of them. We view this inability of reproducing the observed flight and ICT distributions is due to the insufficient description of spatial and temporal movement features. From the analysis results of 10 empirical data sets, we report the following findings, (1) human spatial gathering patterns have a fractal nature, (2) people are likely to visit destinations nearer to their current positions when visiting multiple destinations in succession and (3) each hotspot where people gather together has a lifecycle function in which the number of visitors to that hotspot varies over time and the characteristics of lifecycle functions are strongly correlated with the locations and fractal dimensions of hotspots. In this dissertation, we show that those features are key factors to reproduce heavy-tail flight and ICT distributions that match the distributions observed from empirical data sets. Based on these observations, we propose a mobility model, called Spatio-TEmPoral model (STEP), that contains all the properties mentioned above. With this model, we could generate synthetic heavy-tail flights and ICTs that exactly match those observed from empirical data sets. The contributions of this dissertation are as follows. First, we have collected fine-grained human movement traces using high quality GPS devices in various environments. Five sites are chosen for collecting traces. They are two university campuses (NCSU and KAIST), one metropolitan area (New York City), two theme parks (Disney World and North Carolina state fair). Second, we investigate spatial and temporal movement patterns from the empirical data sets and propose a mobility model that can represent all those patterns. The model is verified using the measured flight and ICT distributions to prove realism. Finally, we have shown the impact of each pattern on the routing performances of DTN.