Non-Linear Offline Time Synchronization

In the design process of communication protocols it is necessary to perform repeated network communication experiments. Each run results in large event logs. The analysis of these logs is crucial to find and to understand unexpected behaviors and design flaws. Intrinsic to network communication these logs suffer from random delays, drop outs, and deviating clocks, which complicate the analysis. Online synchronization protocols may interfere an experiment gravely and are unable to handle delays and drop outs. Offline synchronization approaches based on affine linear clocks using maximum likelihood estimation and least squares estimation are introduced by [3] and [2], respectively. We show that their approaches can be extended to non-linear clocks. The problem leads to a sparse linear program with a well-known structure, which can be readily solved by the interior point method. Under weak assumptions a consistency result is available for the least squares estimation.