Minimization and Placement of Sensors in Structurally Observable Networks

This work concerns with the inference of the network state by monitoring a subset of its nodes. These nodes, which serve as network sensors, can be used to build a distributed monitoring system based on a new network tomography model. First, we present the model in which we can represent the network structure as a linear discrete time invariant dynamical system. Using this model, we define the concept of network structural observability and present an efficient algorithm to minimize the cardinality of the subset of monitoring sensors. As a case study, we use topological data from IPv6 Internet to present some properties of this minimum monitoring subset. To the best of our knowledge, this is the first work to (i) present general properties of sensors placement, (ii) use the proposed model to design distributed monitoring systems, and (iii) illustrate the feasibility of computer networks observability. Moreover, we believe that the theory of network structural observability presented in this paper could significantly benefit the field of network tomography.