Prediction and Planning of Distributed Task Management Using Network Centrality

This work explores the application of complex network centrality metrics as a means to predict the relative location performance of a task manager within a network topology. Our studies examine the problem of optimizing the location of a tasking and data collection node in a master-slave distributed tasking model. Through a series of analytical and emulation experiment-based measurements we examine the correlation between the workflow completion delay and a series of candidate network centrality metrics. Our hypothesis is that centrality based metrics can serve as an engineering guide in planning or managing complex distributed network service deployments. Our initial analytical network experiments are represented by a series of different styles of randomly generated topologies as well as a series of network snapshots from a modeled heterogeneous network topology. Using the baseline analytical model we demonstrate correlation statistics between certain node centrality metrics and the total tasking delay experienced when the task manager is placed at a candidate location. Next, we obtain a set of performance results through emulation experimentation using working distributed coordination software, ad hoc routing, bandwidth restrictions, and network delay within an emulated mesh topology. We perform a series of experiments with varying network diameters and we conclude by reviewing the correlation trends between delay performance prediction and several centrality metrics. We also present a discussion of additional open issues and future work planned.