Efficient algorithms in emerging large-scale networks

Proliferation of today's mobile and social applications has created a variety of logical large-scale networks. These include wireless networks that provide connectivity to billions of mobile devices, road networks that cover thousands of cities, and online social networks that connect more than half a billion people together. As more and more people rely on these applications to communicate and interact, these networks continue to grow rapidly. These large-scale networks raise significant challenges in computation tasks that support the emerging applications. First, due to their scale and complexity, computing tasks that run on these networks are often computational-intensive. Even the most basic problems such as finding the shortest path between two nodes takes hours to solve on a billion-node-graph. Not to mention approximating NP-hard optimization problems. Second, physical resources available to support these computing tasks, such as memory, human power, etc., are heavily constrained. A machine's physical memory can no longer hold an entire network graph, and it is cost-prohibitive to manually collect information to build a road graph. Finally, these networks are dynamically changing, and their applications require up-to-date results. One must design efficient pre-computation so that they can offer up-to-date results without significant overhead. Given these significant challenges in real-time graph computation, conventional solutions are no longer suitable or feasible to accomplish those tasks. We must design a new class of highly scalable algorithms that adapt to continual network dynamics. In this thesis, we investigate efficient algorithm design in three representative large-scale networks: wireless, road, and online social networks. Specifically, we design algorithms that allocate spectrum in wireless networks where devices dynamically share radio spectrum; we design algorithms that build a routable road map from a set of GPS traces collected from vehicles; we also design algorithms that compute approximate shortest paths in online social networks. Our work advances the state of the art by solving the above challenges in several emerging problems and inventing new techniques that potentially inspire future research on related problems.