| Abstract: |
Time synchronization is an important service in many distributed systems. The goal of time synchronization is to provide a common time base, which allows for coordinated actuation and sensor
fusion. This thesis is about guarantees on the quality of time synchronization that can be pro-vided for real systems. Such guarantees can be obtained by measuring the synchroniza-tion quality in
a real system or by using analytical methods and formal system models. In the first part of the thesis, we apply both approaches in the context of the wireless-loudspeakers application. A novel
class of clock synchronization algorithms is intro-duced, which, in contrast to known state-of-the-art algorithms, achieve sufficient syn-chronization quality in a wide range of conditions, that
is in various operation modes of the wireless network and in situations of heavy network traffic. A novel framework for the optimization and comparison of clock-synchronization algorithms is
presented. The framework combines trace-based simulation and evolutionary optimization to provide for a realistic and reproducible evaluation and a fair comparison. The comparison of empiri-cal
and analytical worst-case results allows to argue about the suitability of system mod-els. A novel delay model is proposed that captures the relevant properties of measured delay sequences more
accurately than previously known delay models. In the second part, this thesis also contributes to the understanding of the worst-case syn-chronization quality that can be achieved in large-scale
systems. A novel class of algo-rithms is proposed that has the advantage over known approaches that it is completely local. No global configuration is required and thus a high resilience against
node and link failures is achieved. |
