PROGRESS research projects

RATAD - Reliability and Timing Analysis of Distributed systems

Leader: Hans Hansson Members: Hans Hansson, Thomas Nolte, Christer Norström Lab: Division of Embedded Systems Status: finished Funding: Swedish Foundation for Strategic Research (SSF) via the research programme ARTES, the Swedish Foundation for Knowledge and Competence Development (KK-stiftelsen), LM Ericsson's Research Foundation, and Mälardalen University. Web: Official university web page (in Swedish)

Overview Reliability and Timing Analysis of Distributed systems. --- Traditionally, real-time scheduling theory has targeted hard real-timesystems which are, most of the time, safety critical. Typically, thesesystems can be found as small embedded applications, operated in acontrolled environment. Since these systems are not allowed to have any flaws (in the sence of timliness), analysis thechniques need to be precise. Having precise methods, taking all possible scenarios of execution combinations and execution times into consideration, the analysis becomes very pessimistic which calls for more expensive applications in terms of resources and hardware. However, today moreand more applications have real-time demands, although not assafety-critical as traditional real-time systems, e.g. audio and video transmissions. These new applications open the door to loosen up a bit, thus allowing some deadline misses every now and then. We call this controlled optimism. By using distributions of values instead of worst-case values, designers are given more freedom in terms of arguments for making well founded trade-offs when designing anapplication. We define this as the first problem. Moreover, traditionally scheduling paradigms can be divided into 3 different groups: (1) PRIORITY DRIVEN, (2) TIME DRIVEN, and (3) SHARE DRIVEN. For the Controller Area Network, priority driven scheduling is the natural scheduling method supportedby the CAN protocol, and scheduling analysis methods have been presented. Regarding time driven sceduling, methods have also been implemented. One of those methods also supports priority driven scheduling in combination with time driven scheduling. However, yet so far, share driven scheduling has notbeen implemented. By providing the option of share driven scheduling of the CAN we believe that designers are given more freedom in designing an application. This is our second problem. Therefore, the problems that we try to solve are described in two parts: (1) Probabilistic modelling of thebit-stuffing mechanism of the Controller Area Network (CAN), and (2)Server based scheduling of CAN.

Latest project publications [ Show all publications ] Server-Based Scheduling of the CAN Bus, Thomas Nolte, Mikael Nolin, Hans Hansson, Proceedings of the 9th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA'03), p 169-176, IEEE Industrial Electronics Society, Calouste Gulbenkian Foundation, Lisbon, Portugal, September, 2003 Using Servers to Provide Bandwidth Isolation on the Controller Area Network, Thomas Nolte, Mikael Nolin, Hans Hansson, Proceedings of the 2nd International Workshop on Real-Time Lans in the Internet Age (RTLIA'03) in conjunction with the 15th Euromicro International Conference on Real-Time Systems (ECRTS'03), p 43-46, ISBN 972-8688-12-1, Polytechnic Institute of Porto, Portugal, Editor(s):Instituto Politecnico do Porto, June, 2003 Probabilistic Worst-Case Response-Time Analysis for the Controller Area Network, Thomas Nolte, Hans Hansson, Christer Norström, Proceedings of the 9th IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS'03), p 200-207, IEEE Computer Society, Washington, DC (relocated from Toronto), USA, May, 2003

Results achieved In order to solve the first problem, we try to model the bit-stuffing mechanism in order to retrieve some useful distributions of stuff-bits which could be used instead of the worst-case scenario. By using distributions the level of pessimism is significantly lowered which allows for a greater degree of trade-offs in the sence of timing and reliability. We have written some papers which presents a method to model the stuff-bits as well as some methods to actually decrease the occurence of stuff-bits. For the second problem we propose the usage of server based scheduling techniques, which improves existing thechniques since: (1) Fairness among the messages is guaranteed (i.e., ''misbehaving'' aperiodic processes cannot starve well-behaved processes), and (2) aperiodic messages are not sent ''in the background'' of synchronous messages or in separate time-slots. Instead, aperiodic and periodic messages are jointly scheduled using servers. This substantially facilitates meeting response-time requirements.