工業控制系統中的可擴展性網路路由

Abstract

工業控制系統 (Industrial Control System) 中的網路之擴展性是一個行之有年的研究項目，其目的多為在功能更新或系統有所變動時，能降低重新分配資源之成本。其研究多針對訊息之排程方面的擴展性。而近年來推動的工業4.0中，為滿足客製化或數據分析需求的不定期網路流量開始出現在工業網路中，於是本論文從路由的擴展性切入，探討如何對工控系統提供靜態路由，以滿足未來加入的週期不可預期之流量。我們將網路中的流量區分成兩種，包括週期固定以及不固定者。並針對所有流量提供靜態路由，目標為儘管在不固定周期之流量出現時，其路徑依然能確保所有流量在端到端延遲方面的表現。針對兩種不同之流量，我們分別提出了多種基於啟發式或線性規劃之算法。尤其針對非週期性的流量之路由，引入了環狀路由的概念，藉此降低中央網路控制器的負擔以及轉送表的長度來追求較低的延遲。另外也將這些路由方案套用在實驗案例上，並使用 Omnetpp 模擬器來模擬網路之延遲狀況，藉此檢視所設計之路由算法的成效與理想狀況下之最短延遲之比較，以及在模擬時間結束前是否成功送達所有封包。經過特定方法的搭配，造成的延遲通常不到最佳延遲之兩倍，且成功將所有封包送達的比例在流量較稀疏的案例中到達九成，在流量較密集的案例中亦有七成。

The extensibility of networks in real-time systems has been a longstanding research topic, aiming to reduce the cost of resource reallocation when there are functional updates or system changes. Much of this research focuses on the extensibility of message scheduling. In recent years, with the advancement of Industry 4.0, intermittent network traffic aimed at meeting customization or data analysis needs has begun to appear in industrial networks. Therefore, this thesis approaches the issue from the perspective of routing extensibility and explores how to provide static routing for industrial control systems to accommodate the unpredictable periodic traffic that may be introduced in the future. We categorize traffic in the network into two types: periodic and non-periodic. Static routing is provided for all traffic, with the goal of ensuring that even when non-periodic traffic appears, the paths still guarantee performance in terms of end-to-end delay for all traffic. For the two different types of traffic, we propose multiple algorithms based on heuristics or linear programming. Especially for non-periodic traffic routing, the concept of ring routing is introduced to reduce the burden on the central network controller and the length of forwarding tables to achieve lower delay. Additionally, these routing schemes are applied to experimental cases, and the Omnet++ simulator is used to simulate network delay conditions, examining the effectiveness of the designed routing algorithms compared to the ideal shortest delay condition, and whether all packets are successfully delivered before the end of simulation time. Through specific method combinations, the resulting delay is typically less than twice the best delay, and the proportion of successfully delivered packets reaches ninety percent in cases of sparse traffic and seventy percent in cases of dense traffic.