Pluntke, C; (2014) Modelling users in networks with path choice: four studies in telecommunications and transit. Doctoral thesis , UCL (University College London).

PDF Christopher_David_Pluntke_thesis_final.pdf Download (2MB)

Abstract

Networks of interacting users arise in many important modelling applications. Commuters interact with each other and form traffic jams during peak-time. Network protocols are users in a communication network that control sending rate and server choice. When protocols send with too high rates, network links get overloaded resulting in lost data and high delays. Although these two example users seem very different, they are similar on a conceptual modelling level. Accurate user models are essential to study complex interactions in networks. The behaviour of a user with access to different paths in a network can be modelled as an optimisation problem. Users who choose paths with the highest utility are common in many different application areas, for example road traffic, Internet protocol modelling, and general societal networks, i.e. networks of humans in everyday life. Optimisation-based user models are also attractive from the perspective of a modeller since they often allow the derivation of insights about the behaviour of the entire system by only describing a user model. The aim of this thesis is to show, in four practical studies from telecommunications and transit networks, where optimisation-based models have limitations when modelling users with path choice. We study users who have access to a limited number of paths in large scale data centers and investigate how many paths per user are realistically needed in order to get high throughput in the network. In multimedia streaming, we study a protocol that streams data on multiple paths and path properties matter. We also investigate complex energy models for data interfaces on mobile phones and evaluate how to switch interfaces to save energy. Finally, we analyse a long-term data set from 20,000 transit commuters and give insights on how they change their travel behaviour in response to incentives and targeted offers. We use tools from optimisation, simulation, and statistics to evaluate the four studies and point out problems we faced when modelling and implementing the system. The findings of this thesis indicate where user models need to be extended in order to be of practical use. The results can serve as a guide towards better user models for future modelling applications.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item