Georgia Institute of Technology
How New Internet Mechanisms Reshape Deployment and Measurement
Abstract
dc:description.abstractInternet infrastructure has evolved significantly since its inception. However, monitoring and characterizing Internet infrastructure, such as performance and deployment of various technologies, presents open problems with significant implications. As IPv6 deployment increases, measurement techniques and analysis in IPv4 leave behind a large fraction of the Internet. From this arise other questions of Internet infrastructure usage, such as the state of IP sharing deployment and how bots powering modern AI services are deployed. This dissertation presents empirical methods to characterize the new state of these phenomena at scale. First, we discuss work that evaluates expectations about IP-based infrastructure. By analyzing data from the governing organizations that allocate Internet number resources, we find that such heuristics used in IPv4 measurement, such as typical prefix sizes, do not exist in IPv6. Rather, deployment strategies are heterogeneous. We then discuss work that measures a transition mechanism that bridges connections together, allowing an IPv6-connected device to access resources hosted in IPv4. While we measure from end-user devices and over 1 million open resolvers, we find few measurable deployments, contradicting anecdotal reports from operators. Thus, this work highlights a gap in our measurement abilities, obfuscating a key dimension of IP connectivity. Next, we shift our focus to how the IP version may affect end-user network performance. Using a residential measurement platform, we systematically execute and compare multiple speed test software in each IP version. We identify several systematic factors that correlate with significant differences in performance, including providers and speed test measurement infrastructure itself. This work highlights an important, yet frequently forgotten, network characteristic and calls for future work to rethink analysis and measurement. We then analyze large-scale network deployment characteristics from the perspective of a global Content Delivery Network (CDN). We present a method to detect massive IP sharing deployment (e.g., Carrier-Grade NAT) that solely relies on high-level traffic patterns. Our findings paint a stark picture of the Internet; over 40% of IPv4 traffic emerges from less than 2% of active addresses. We additionally find surprising deployments of large-scale IPv6 sharing, accounting for over 7% of IPv6 traffic in North America. We then shift our focus to another widely deployed mechanism: LLM-based AI services. Here, we provide an unprecedented view into the infrastructure sending traffic from the bots and crawlers supporting these services, providing granular transparency that can enable future work in understanding this new ecosystem. Ultimately, this dissertation shows that IPv6 adoption and LLM-based AI services have fundamentally changed the deployment of IP-based infrastructure and end-user browsing experiences. Thus, we need to empirically reevaluate IP address usage, IP connectivity, access network performance, and what infrastructure is used in content retrieval, to support an observable, well-functioning, and usable Internet.
Degree
thesis:*- Name thesis:degree_name
- Computer Science, PhD
- Department thesis:degree_department
- Computer Science
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Hsu, Amanda Catherine
- Advisor dc:contributor.advisor
-
- Li, Frank
- Committee members dc:contributor.committeemember
-
- Zegura, Ellen
- Testart, Cecilia
- Pearce, Paul
- Feamster, Nick
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/82211