UC Berkeley

The state of security across the Internet is poor, and it has been so since the advent of the modern Internet. While the research community has made tremendous progress over the years in learning how to design and build secure computer systems, network protocols, and algorithms, we are far from a world where we can truly trust the security of deployed Internet systems. In reality, we may never reach such a world. Security concerns continue to be identified at scale through-out the software ecosystem, with thousands of vulnerabilities discovered each year. Meanwhile, attacks have become ever more frequent and consequential.

As Internet systems will continue to be inevitably affected by newly found security concerns, the research community must develop more effective ways to remedy these issues. To that end, in this dissertation, we conduct extensive empirical measurements to understand how remediation occurs in practice for Internet systems, and explore methods for spurring improved remediation behavior. This dissertation provides a treatment of the complete remediation life cycle, investigating the creation, dissemination, and deployment of remedies. We start by focusing on security patches that address vulnerabilities, and analyze at scale their creation process, characteristics of the resulting fixes, and how these impact vulnerability remediation. We then investigate and systematize how administrators of Internet systems deploy software updates which patch vulnerabilities across the many machines they manage on behalf of organizations. Finally, we conduct the first systematic exploration of Internet-scale outreach efforts to disseminate information about security concerns and their remedies to system administrators, with an aim of driving their remediation decisions. Our results show that such outreach campaigns can effectively galvanize positive reactions.

Improving remediation, particularly at scale, is challenging, as the problem space exhibits many dimensions beyond traditional computer technical considerations, including human, social, organizational, economic, and policy facets. To make meaningful progress, this work uses a diversity of empirical methods, from software data mining to user studies to Internet-wide network measurements, to systematically collect and evaluate large-scale datasets. Ultimately, this dissertation establishes broad empirical grounding on security remediation in practice today, as well as new approaches for improved remediation at an Internet scale.