Implementation and Evaluation of Multipath TCP (MPTCP) Schedulers For Reliable and Low-latency Car-to-Cloud Communication

Recent developments in the automotive and communication sectors enabled vehicles to operate as a highly mobile sensor platform. As the connected car concept continues to gain traction, Car-to-Cloud communication can support the transmission of collected data to the cloud, e.g., dynamic collection of new map information or localized weather forecasts. In addition to low latency and reliability, the data transmission requires a stable connection between a mobile vehicle and the cloud resources. Compared with only a single LTE connection such as regular TCP, leveraging multiple LTE interfaces (e.g. multipath protocols) can dramatically improve the transmission speed. However, MPTCP lacks the support to prioritize one path over another. In case the network providers have different costs, it may cause undesired network usage of the most expensive connection, resulting in a higher overall transmission cost. The thesis focuses on the implementation in the Linux kernel and evaluation through MiniNet of MultiPath Deadline and Preference-Aware (MP-DPA), a multipath framework inspired to MP-DASH with the awareness of network interface preferences. The basic idea behind MP-DPA is to strategically schedule packets' delivery while maintaining a reliable transmission and satisfying interface preferences. The measurements, with symmetric and asymmetric static bandwidth conditions, define the effective advantages and limitations of MP-DPA in terms of transmission cost, which can be adapted by tuning the so-called deadline sensitivity α. The extensive trace-driven studies in highway, rural, and city locations in the North Rhine-Westphalia state in Germany, validate MP-DPA improvements also in a dynamic context. For simulation purposes, are considered two German network providers: O2 (Telefonica) assigned to the preferred interface and Telekom, considered more expensive, attributed to the non-preferred one. MP-DPA is investigated for bulk upload, a car-like transmission, and stream-like upload, typical of video streaming contexts. The results show that, compared with the off-the-shelf MPTCP, MP-DPA minimizes the non-preferred LTE / Telekom usage without compromising the reliability. The driver position influences the MP-DPA effectiveness: in a rural location, where the network coverage is poor, Telekom usage is reduced by 6% and 4% compared to the baseline MPTCP, with bulk and streaming upload, respectively. The best performances are obtained in urban locations where savings percentages are increased by up to 16%. The thesis concludes with an overview of the results obtained through realistic trace-driven experiments and an outlook on possible future works on MP-DPA features and implementation modifications.