Modular Performance Analysis of Distributed Embedded Systems

Abstract

Embedded computer systems are getting increasingly distributed. This can not only be seen on a small scale, e.g. in terms of multiprocessors on a chip, but also in terms of embedded systems that are connected via various communication networks. Whereas classical methods from the worst case timing analysis and real-time community focus on single resources, new models and methods need to be developed that enable the design and analysis of systems that guarantee end-to-end properties.

The talk covers a new class of methods based on real-time calculus. They can be considered as a deterministic variant of queuing theory and allow for (a) bursty input events and event streams, (b) heterogeneous composition of scheduling methods (EDF, FP, TDMA, WFQ, ...), (c) distributed computation and communication resources (d) detailed modeling of event stream correlations and resource behavior and (d) hard worst case bounds. In addition, the methods have been combined with formal assume/guarantee interfaces. Besides introducing the basic models and methods, some application studies are covered also.

It appears that this class of new methods provide a major step towards the analysis and design of predictable distributed systems.