Challenges in streaming graph analysis

The volume of streaming data for cyber analysis is increasing at a rate much greater than any organization's ability to hire human analysts. As a preliminary step to automating significant portions of analysis workload, we consider the problem of modeling cyber data. Since the latter tends to be relational in nature, graphs are a natural abstraction. This motivates future research into efficient algorithms for fundamental graph problems in a high-volume, streaming environment. Algorithms designed using current theoretical models for streaming graph algorithms are not directly suitable for operations. In this talk, we propose a new streaming model that can be implemented on a parallel system with extremely simple topology. Our model assumes an infinite stream which must be analyzed using finite resources. We illustrate the associated challenges by giving an algorithm for maintaining and querying the connected components of a graph in which edges may be expired periodically.