| Outages and loss of connectivity in the Internet can have a significant impact on businesses and users. While the Internet was designed to recover from simple failures, there are numerous examples of accidents and attacks over the past two decades that have resulted in large-scale loss of connectivity. This suggests that standard mechanisms to ensure Internet robustness require renewed consideration. Toward the goal of improving Internet robustness, this dissertation has two themes: (i) understanding the structural complexity and inherent risks in the Internet---two factors which makes the Internet failure-prone, and (ii) building scalable, robust and easy-to-deploy systems to enhance network robustness.;In this dissertation, we first investigate how a comprehensive understanding of structural complexity and inherent risk are critical for robust Internet design and operation. A key challenge in this regard is that Internet is in a constant state of flux. To overcome this challenge, we take a bottom-up approach and build Internet Atlas, which is a comprehensive repository of the physical Internet. The main goal of Internet Atlas is to develop a geographically accurate representation of the Internet's physical interconnection infrastructure map to understand the structural complexity. We also develop a new probing heuristic called POPsicle to broadly identify Internet infrastructure that has a fixed geographic location such as Point-of-presences (POPs), Internet exchange points (IXPs), datacenters and other kinds of hosting facilities.;One of the most striking characteristics of the constructed map is a significant amount of observed infrastructure sharing, which in turn aggravates the risks inherent on topology. Such infrastructure sharing is the result of a common practice among many of the existing Internet service providers (ISPs) to deploy their fiber in jointly-used and previously installed conduits and is dictated by simple economics---substantial cost savings, among other objectives, as compared to deploying fiber in newly constructed conduits. By considering different metrics for measuring the risks associated with infrastructure sharing, we examine the presence of high-risk links in the existing infrastructure, both from a connectivity and usage perspective.;Given the understanding of structural complexity, we next develop systems that take advantage of emerging technology to satisfy ISP objectives and to minimize shared risks. First, we create a decision support framework that uses geo-based multi-objective optimization to identify target areas with the highest concentration of un/underserved users at the the lowest cost to service providers for network infrastructure deployment. Second, we propose a system called GreyFiber, which provides a means to offer easy and cost-effective access to unused fiber-optic paths between participating endpoints on demand based on market economics, for arbitrary durations, and possibly with industry-specific performance guarantees. |
