兩種時間同步化協定的模型建立與比較

Abstract

為了推論出正確的應用程式語意，感測器網路應用程式需要將由分散的感測器節點回報的觀察數據精確地標上時間。由於每個節點各自的時鐘可能由於時鐘飄移而不同步化，因此需要一個時間同步化的協定使所有節點的時鐘跟一個參考時鐘同步。本篇論文提供「時鐘同步」與「事件同步」兩種時間同步化協定的效能評估與比較。它們最大的差異在於「時鐘同步」同步化所有節點各自的時鐘，而「事件同步」同步化每個節點與接收端對於事件的產生時間。雖然這兩種協定在應用程式使用上有各自不同的限制，它們仍然可以在共有的應用程式領域裡作比較。本篇論文在不同網路規模、節點移動程度、及封包交通流量下評估這兩種協定。為了了解這兩種時間同步化協定的優劣取捨，我們推導出它們效能的分析模型並跑模擬及真實實驗以測量這些變數的影響。實驗及模擬結果，還有分析模型都顯示: (1)「事件同步」比「時鐘同步」提供更好的精確度。(2)在節點移動程度高的情況下，「時鐘同步」可能比「事件同步」達到更好的精確度。(3)在封包交同流量大的情形下，「時鐘同步」有較好的擴充性。我們更導出一個在不同網路規模、封包交通流量、及應用程式需求下，能指出如何選取最佳時間同步化協定的選取指南。

To infer correctly application semantics, sensor network applications often need accurate times on observations that are reported from distributed sensor nodes. Since the nodes' local clocks can go out-of-sync due to clock drifts, a networked time synchronization protocol is needed to synchronize their clocks to a reference clock. This paper provides performance modeling and comparison between two time synchronization protocols: TPSN clock synchronization (clock-sync) and TSS event synchronization (event-sync). Their main difference is that the TPSN clock-sync synchronizes all nodes' local clocks to a global reference clock, whereas TSS event-sync synchronizes events' generation times from different local nodes to their sink nodes' clocks. Although these two time synchronization protocols have their respective limitations in application scenarios, they are comparable in that they also share a large domain with none of these limitations. This paper evaluates these two protocols by considering different ad-hoc network sizes, node mobility levels, and traffic volumes. In order to fully understand the tradeoffs between these two time synchronization protocols, we have derived analytical models on their performances and conducted simulations and real experiments to measure the impact of these variables. The experimental results, simulation results, and analytical models all show that (1) event-sync provides much better accuracy than clock-sync, (2) under very high node mobility level, clock-sync may achieve better accuracy than event-sync, and (3) under increasing traffic volume clock-sync scales better. A selection guideline is derived showing how to choose the optimal class of time synchronization protocols under different sensor network dynamics, traffic dynamics, and application requirements.