Quality of Encrypted DNS Resolvers

Recently, many browser vendors have deployed DNS-over-HTTPS (DoH) to encrypt DNS queries in response to growing privacy risks. Most browsers select between only a handful of resolvers despite 100+ resolvers existing. This allows for those mainstream resolvers to sit on heaps of user data, simply shifting the original privacy problems to a different set of third parties. For this reason, we study the wider set of encrypted DoH resolvers to better understand the viability of resolvers beyond the rather popular ones that vendors select from.

We explored the performance of a large group of DoH resolvers through two tools. With the first tool, we measured DNS query response times and ping times from vantage points in North America, Asia, and Europe. For the second tool, we added our measurement harness to an existing network measurement platform, collecting data from 20+ raspberry pi devices in home networks across the Chicago area.

For this project, I used MaxMind’s GeoIP services, DIG, AWS, Matplotlib, and Pandas for data parsing.

User Understanding of Encrypted DNS Interfaces

This project was accepted to the 32nd USENIX Symposium. I presented our paper at USENIX and at IETF 117.

The encrypted DNS ecosystem is increasingly becoming a place in which various internet stakeholders fight for control through technical protocol design. Many browsers provide choices of only a few common resolvers, creating privacy risks because it allows for a handful of mainstream resolvers to sit on heaps of user data. This consolidation creates a power imbalance which has implications for security and performance. Since humans are impacted the most by browser configurations of DNS settings, we studied their awareness and knowledge.

Through a two part survey via Prolific, we considered the Opera, Edge, Firefox, Chrome, Brave, and Android OS encrypted DNS interfaces. After analyzing participants’ responses, we provided recommendations as to how default settings should be designed, the additional information that users should be provided, and how privacy risks can be ameliorated.

For this project, I used Prolific, Qualtrics, Matplotlib, and Pandas.

The Digital Divide across Chicago Neighborhoods

This paper was accepted and presented at TPRC 2022.

There has been discourse around two separate topics—benchmarking Internet performance standards and Internet equity—but methods to address these subjects have been centered on metrics that don’t adequately convey user experience. Further, samples frequently don’t accurately represent populations and are subsequently not suitable for answering questions about equity.

We considered the types of sampling method and metrics that can be used for benchmarking Internet service provider performance. We also highlighted data from a sample across Chicago that is specifically designed to answer questions of equity, studying how Internet performance in a specific high-income neighborhood compares with Internet performance in a specific low-income neighborhood.

For this project, I used Influx, Grafana, Pandas, and Matplotlib.

WiFi Bottlenecks in Home Networks

As broadband Internet speeds increase, WiFi networks are increasingly becoming access bottlenecks. In many cases, bottlenecks lead to the actual speeds which households have to be significantlylower than the Internet speed they buy from their ISP.

We studied the extent to which this bottleneck exists by analyzing home network measurements in 44 households. We continuously measured the user’s Internet performance from a wireless client inside the home network (e.g. the cable modem) and from a wired point of access to the user’s ISP.

For this project, I analyzed data using Pandas and Matplotlib.