Awareness-Driven Phase Transitions in Very Large Scale Distributed Systems

Recent research in the field of complex networks has shown that - beyond microscopic structural qualities - global statistical parameters are sufficient to describe a surprising number of their macroscopic properties. This article argues that such statistical parameters can be monitored by nodes in a decentralized and efficient way. The so achieved awareness of a network's global parameters can be used by nodes for actively influencing them to optimize relevant characteristics of the overall network. For such an adaptation, the network-analogy of "phase transitions" in physical systems can be used. In this article the general concept of such an awareness-driven statistical adaptation is presented using power law networks as an example. For this important class of networks practical algorithms are introduced: Based on recent advances in reliable power law fitting, a gossip scheme has been developed which is suitable to make individual nodes aware of a power law network's critical exponent. In order to influence this parameter, decentralized reconnection rules are presented. The combination of both strategies facilitates a feedback control of large scale systems' emergent power law properties.