A Survey of MARS

Providing a reliable service in a continuously running real-time system is problematic. MARS (MAintainable Real-time System) o ers correct and timely responsiveness by design and ease of maintenance by design of a new fault-tolerant distributed real-time system with active redundancy. Several unconventional techniques are implemented in the operating system of MARS. It is based on a time rigid scheduling, it avoids interrupts and has no dynamic memory allocation. All operating system concepts that may lead to an unexpected run time delay or deadlock are kept out of MARS. Contrary, MARS has new important services within its operating system. A fault-tolerant global time scale with a precision of a few microseconds allows the synchronous execution of distributed cooperating tasks. Tasks communicate through standardized messages with limited life time. The reliable message transport service tolerates transient message losses using redundant datagrams. MARS strongest feature is to interpret each message as a state value, i. e. a newer value updates an older one. This gives notable advantages for the operating system and totally avoids queuing of events. Active redundant components produce the same messages. Due to the state semantics of messages redundant messages are masked out easily. State messages can be read arbitrarily often, but they never can be consumed. Thus multiple or redundant tasks may read the same message. Scheduling in MARS is a combination of time rigid CPU and message bus scheduling. Periodic tasks can be pre-scheduled such that their CPU slot will nish just when access to the communication media will be granted. Consequently, all transactions can be speci ed and their processing time can be checked during the system's design phase. Several methods for fault-detection are included in MARS to achieve a failsilent behavior of system components. All these features of MARS have been implemented and tested experimentally at the Technical University in Vienna. We strongly believe that such a deterministic system design is a feasible technique for engineering distributed hard real-time computers and applications.