Open access
Author
Date
2023Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
In the Internet, billions of endpoints exchange data, and thereby create a global traffic pattern, characterized by a time-variant distribution of traffic volume over inter-domain paths. This global traffic pattern is increasingly reshaped by two mechanisms for data-traffic control that have experienced real-world deployment in recent years. First, end-host path selection allows end-hosts to allocate traffic on inter-domain paths that are under-utilized or particularly suitable for a given application. Second, modern congestion-control algorithms (CCAs) promise to simultaneously achieve high utilization and low congestion, namely with a latency-sensitive approach that can withstand competition from legacy algorithms.
However, despite extensive research into end-host path selection and modern congestion control, the research community does not yet completely understand how these two control mechanisms will affect the Internet traffic pattern under global deployment. To further such an understanding, this thesis follows a model-based approach, i.e., it describes end-host path selection and congestion-control algorithms in a mathematical fashion, and leverages these descriptions to investigate the effects of the emerging mechanisms. Thanks to such formal representation, the model-based analysis in this thesis can rely on diverse mathematical tools, including game theory and control theory, and therefore provide provable guarantees.
The model-based analysis in this thesis is separated into two parts, both representing aspects of the intricate dynamics in Internet traffic:
Part I: Network performance. In the first part of this thesis, I find that load-adaptive path selection by selfish end-hosts admits near-optimal equilibria, but may not converge to these equilibrium distributions. Instead, this path selection may cause oscillation, which demonstrably hampers network performance, but which can be suppressed by network operators. Moreover, I present a highly accurate fluid model of the BBR congestion-control algorithm, and prove that BBR is stable in homogeneous settings, but unstable in competition with loss-based CCAs.
Part II: ISP economics. In the second thesis part, I discover that end-host path selection alters the cooperation among ISPs, by allowing novel interconnection agreements that substantially boost path diversity, and the competition among ISPs, by motivating investments that improve transmission quality.
In summary, the model-based approach in this thesis yields numerous complex insights, and thus contributes towards understanding the future Internet. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000650605Publication status
publishedExternal links
Search print copy at ETH Library
Contributors
Examiner: Perrig, Adrian
Examiner: Vanbever, Laurent
Examiner: Schmid, Stefan
Examiner: Schapira, Michael
Publisher
ETH ZurichSubject
Path selection; Congestion control; Network performance; Internet economicsOrganisational unit
03975 - Perrig, Adrian / Perrig, Adrian
More
Show all metadata
ETH Bibliography
yes
Altmetrics