Dynamically Evolving, Large Scale Information Systems
Growing in a seemingly uncontrolled manner, information systems such as the Internet, networks of literature, of phone calls or of scientific collaborations or peer-to-peer networks cannot any more be developed, managed and maintained with traditional techniques. The reasons for this phenomenon are obvious, global networking und the omnipresence of technical media result in sizes of several million users or interconnected machines, alongside the appropriate dynamics. For such systems we are forced to abandon the goal of global optimality. Instead, we have to focus our efforts on finding scalable, self-regulating and self-repairing mechanisms, which are both able to adapt to changes in their environment and to maintain the system in a feasible state while avoiding instability.
The main target of this project is the development of methods, techniques and tools by means of interdisciplinary efforts in the fields of computer science, physics, biology, and economics, that are capable of dealing with the challenges of such systems. The Karlsruhe group is part of subproject 1 of DELIS, which focuses on “Monitoring, Visualizing, and Analyzing Large Dynamically Evolving Information Systems”. The goal of this subproject is the development of concepts, theoretical foundations, algorithms, tools, prototypes and software platforms which help to grasp and comprehend these information systems and to represent them in an appropriately designed analytic visualization.
Beside the topic of graph clustering - the automated identification of natural groups in networks - we concerned ourselves with analytic visualizations, in the past and final year of the project DELIS. Visual analysis of networks has changed substantially in the recent past. The availability of large-scale digitized networks, powerful hardware and the discovery of a number of defining characteristics of complex networks have resulted in visualizations nowadays being created on a new level. Such visualizations are called fingerprints, they incorporate analytic properties of the network into the layout and thus exhibit these in the context of the network structure. A visualization technique of this kind can, on the one hand, serve as a tool for a traditional network analysis, but has, on the other hand, an exploratory character and is capable of revealing properties present in a network, either those that are expected or others which might not have been foreseen. As a pathbreaking technique for compiling fingerprints we have developed LunarVis. The visualization algorithm of LunarVis is specialized on displaying a user-defined partition of the set of nodes together with analytic properties such as node centrality or edge weights in the context of the neighborhood structure of nodes.