Predictable high-speed communications for distributed real-time systems

Sammanfattning: The ubiquitous use of computers in modern society is rapidly changing our way of life. In many existing and emerging applications the timely response to events is of utmost importance. Consider for instance a modern vehicle with embedded computers that must react to driver commands within well specified timing tolerances. In designing this and many other type of systems there is a need for methods that can assist in assessing proper functionality and timing. Since many of the involved computers need to communicate with other computers to perform their functions, there is a specific need to analyse communications between computers. In particular, methods to calculate the delay of a message sent through a communications network is crucial in the design and validation of distributed hard real-time systems, such as the above mentioned vehicle computer system. The use of such systems is increasing, due to the increased performance and reliability they can offer, as well as the increasing need to control physically distributed equipment. As a consequence, methods to provide timeliness guarantees become more important. To assist in providing guarantees of timely delivery of messages, this thesis presents a method to determine the worst case response-time and amount of memory needed for messages sent over Asynchronous Transfer Mode (ATM) networks. An associated priority assignment algorithm, essential for achieving efficient sharing of network resources, is also presented. The methods are intended for admission control of hard real-time traffic, i.e. they can be used to determine if timing requirements will be met or not. In a comprehensive evaluation, we compare our method with other methods thatprovide similar guarantees, and show that our response-time analysis enables higher utilisation of network resources compared to the other methods. We show that this success is a result of the high precision in our analysis, combined with the good performance of the priority assignment algorithm. In addition, we show the versatility of our method by expanding it to handle more complex traffic patterns which occur in compressed multimedia traffic, and we also present methods improving the efficiency of our analysis.