Evaluating Congestion Control Algorithms and Predictive Quality of Service

Christian Leonard Vielhaus

AutorIn

Christian Leonard Vielhaus

Titel

Evaluating Congestion Control Algorithms and Predictive Quality of Service

Zitierfähige Url:

https://nbn-resolving.org/urn:nbn:de:bsz:14-qucosa2-1039680

Erstveröffentlichung

2026

Datum der Einreichung

02.06.2025

Datum der Verteidigung

20.11.2025

Abstract (EN)

The Internet is a cornerstone of modern civilization, enabling worldwide communication and information exchange for millions of daily users. Since the Internet’s early days, Congestion Control Algorithms (CCAs) have played a crucial role in ensuring efficient, reliable, and fair data transmissions. For over three decades, hundreds of CCAs have been proposed and evaluated. These evaluations are as diverse as the CCAs themselves using various scenarios and metrics to assess algorithmic performance. To this day, however, there is no consensus on how to evaluate CCAs comprehensively. Furthermore, there is a lack of transparency and reproducibility of evaluation results. This thesis proposes ccperf, an evaluation framework that aims to provide a unified methodology for evaluating CCAs in a thorough, fair, and reproducible manner. To achieve this aim, ccperf is designed to encompass a broad set of evaluation scenarios and metrics commonly used in the literature. The capabilities of ccperf are demonstrated by comparatively evaluating selected state-of-the-art CCAs to identify their strengths and weaknesses and provide insights into design trade-offs. As a contribution to the community, ccperf is an open-source framework based on both Linux emulations and ns-3 simulations. Other researchers may use the framework to evaluate existing CCAs or to support the prototyping and evaluation of new CCAs throughout their development. CCAs are important for the Quality of Service (QoS) of communication networks. In the realm of wireless networks, the QoS often varies over time, and the best effort approach of communication networks provides no guarantees. However, many applications demand a certain QoS to function properly. To bridge this gap, predictive Quality of Service (pQoS) algorithms can be used to predict the QoS of a communication network. This thesis proposes pQoSc, an open-source pQoS library that predicts the QoS in cellular networks with the help of Machine Learning (ML) algorithms. The focus is on predicting two selected QoS metrics: the maximum achievable throughput and handover timings. The QoS metrics are predicted based on datasets stemming from real-world measurement campaigns and cellular network simulations. The simulations reproduce a measurement campaign in Berlin, Germany, to generate synthetic datasets from a virtual reconstruction of a real-world scenario named vBerlinV2N. Using these datasets, the QoS metrics are predicted with high accuracy and pQoSc provides further insights into pQoS wth ML algorithms for cellular networks.

Freie Schlagwörter (EN)

Networks, Congestion Control, Quality of Service, Machine Learning

Klassifikation (DDC)

621

Klassifikation (RVK)

ZN 6560

GutachterIn