The power of locality: Exploring the limits of randomness in distributed computing

Many modern systems are built on top of large-scale networks like the Internet. This article provides an overview of a dissertation [29] that addresses the complexity of classic graph problems like the vertex coloring problem in such networks. It has been known for a long time that randomization helps significantly in solving many of these problems, whereas the best known deterministic algorithms have been exponentially slower. In the first part of the dissertation we use a complexity theoretic approach to show that several problems are complete in the following sense: An efficient deterministic algorithm for any complete problem would imply an efficient algorithm for all problems that can be solved efficiently with a randomized algorithm. Among the complete problems is a rudimentary looking graph coloring problem that can be solved by a randomized algorithm without any communication. In further parts of the dissertation we develop efficient distributed algorithms for several problems where the most important problems are distributed versions of integer linear programs, the vertex coloring problem and the edge coloring problem. We also prove a lower bound on the runtime of any deterministic algorithm that solves the vertex coloring problem in a weak variant of the standard model of the area.