Locality, scheduling, and selfishness: Algorithmic foundations of highly decentralized networks

Abstract

Large-scale and highly decentralized networks such as the Internet have emerged as arguably the most complex computer systems. The size of such networks, as well as their dynamic, socio-economic, and often wireless nature brings about a large variety of challenging problems. Achieving efficient and provably robust algorithmic solutions for these problems, in turn, necessitates the development and employment of novel techniques and methods. This dissertation studies three major concepts that stand out as being of particular importance and interest in this context: locality, wireless communication, and selfishness. In large-scale networks, gathering information about the whole network topology is either too resource consuming or simply impossible due to mobility, dynamics, or churn. Hence, no node is typically able to collect or maintain a global state about the network and each node has to base its decision on local information only. This discrepancy between the need to achieve global efficiency and the limitation to local knowledge motivates the study of local algorithms and local computation. The first part of this dissertation investigates the possibilities and limitations of local computation in the classic message passing model of distributed computing. We present near-tight upper and lower bounds on the achievable trade-off between the amount of local knowledge and the resulting global solution for a variety of fundamental network problems. The integration of wireless devices into the Internet, and the advent of wireless multi-hop networks such as ad hoc and sensor networks poses numerous algorithmic challenges. This dissertation investigates the distributed complexity of computing network coordination structures such as clusterings and colorings in models that closely capture unstructured wireless multi-hop networks. We then define and study the scheduling complexity of wireless networks in a physical model of wireless communication. This measure describes the theoretically achievable performance of any scheduling protocol and allows to characterize and analytically evaluate existing protocols from a worst case perspective. Hosts in the Internet or peers in a peer-to-peer network are typically governed by socio-economic agents whose main interest is not the benevolent optimization of the network’s entirety, but rather the maximization of their own benefit. In other words, participating agents may be selfish, rather than acting in a coordinated manner that optimizes social welfare. The final part of this thesis analyzes the impact of selfish and potentially malicious behavior on the efficiency of peer-to-peer and other decentralized computer networks.