DFG project: A modular framework for modeling mobility behavior in wireless networks considering varying granularity of data (2016-2019).
An essential issue in the stochastic modeling of computer networks is the adequate and realistic modeling of the loads that occur. With the increasing spread of mobile devices, the performance evaluation of wireless networks plays an increasingly important role. For a realistic modeling of the loads, it is necessary to represent the mobility behavior of the users appropriately.
When generating these mobility patterns, one can basically distinguish between approaches that generate the patterns completely artificially and approaches that take real data as a basis for generation. The generation of mobility models based on real data is significantly more complex, but the generated motion patterns prove to be more realistic than artificially generated ones. However, one difficulty in creating these mobility models arises from the granularity of the available data. For example, data automatically recorded at the access points of a WLAN do not contain precise information about movement patterns, but only allow users to be assigned to the respective access points. In contrast, data sets have been obtained from smaller user groups in the past that allow a more precise determination of the respective whereabouts. Depending on the granularity of the data and also on the goal of the analysis of the model, different approaches to modeling mobility behavior have been developed in the past, but these approaches cannot be easily transferred to other study scenarios.
Therefore, the aim of this project is to develop a modular framework that takes into account the varying granularity of the data and different analysis goals. The mobility models to be developed should represent the movement patterns at different levels of abstraction and be modularly extensible in order to be able to increase the level of detail of the model in a simple way. In this way, the framework to be developed provides a basis with which the effects of different levels of modeling abstraction on performance measures can be easily compared.