Distributed Network Tomography Applied to Stochastic Delay Profile Estimation

Abstract

In this paper is shown, how delay properties of the edges of a network with stochastic properties can be estimated cooperatively by individual nodes that retain the delay profiles of the entire network. The proposed algorithm adopts null-space projection-based consensus among agents to find individual entries from a set of arbitrary sum-cumulative entities associated with graph edges (e.g., delays associated with edges) based on sums over the network paths. The local estimates of delay profile are estimated using Least Squares (LS). A modified, tailored, iterative consensus algorithm is then employed to distribute information among the neighbors. The distributed network tomography is compared to the conventional centralized solution and also to iterative solvers based on Cimmino, CAV, and Landweber methods applied in a distributed manner.