Bookshelf

European Train Control System (ETCS) on-board units are safety-critical systems whose reliability plays a vital role in the integrity of railway operation. While redundancy is a typical method for increasing fault tolerance and hence reliability and safety, it also elevates synchronization and runtime complexity. This thesis proposes the usage of DDS - a real-time capable machine-to-machine communication standard following the Data-Centric Publish-Subscribe (DCPS) pattern - to deal with the added complexity. It further provides background and implementation of a redundant and fault-tolerant ETCS on-board system using real-time DCPS communication and consensus-based voting. Besides data-centric implementations of behavioral concepts, such as a leader election and decision-making algorithm, the work contributes practical solutions to global state and recovery subjects in distributed systems. Functionality, safety, and reliability of the implementation are evaluated using a subset of ETCS and a cluster of four Raspberry Pi-based Programmable Logic Controllers (PLCs) within a simulated environment. The results show that (1) the redundant system facilitates real-time computation with high network throughput (up to 77 Mbit/s) and fast consensus-building (µ=6.5 ms; σ=1.3 ms). (2) Consensus-based DCPS communication increases reliability and enables fast system recovery (µ=5.47 ms; σ=0.77 ms upon recognizing a fault). (3) A subset of DDS is al- ready sufficient to accomplish reliable and safe ETCS on-board units. Thereby, this work facilitates the development of safe and cost-efficient on-board systems for future ETCS applications.