Abstract: The paper addresses a novel method for performance analysis of
distributed realtime systems with complex, and especially cyclic data flow
graphs. The presented method is based on Network Calculus principles, where
flow and service constraint functions are used to bound data flows and
processing resources. In acyclic systems flow constraints may be propagated
through a system of service models along data flow paths, whereas service
constraints are propagated along paths of priority order. For cyclic data flows
however constraint propagation becomes equivalently cyclic leading to flow and
service constraints implicitely given by a fix point equation in a space of
constraint functions. In this paper a method denoted CyNC for obtaining a well
defined solution to that problem is presented along with a theoretical
justification of the method as well as comparative results for CyNC and
alternative methods on a relevant example. The method is implemented in a
prototype tool also denoted CyNC providing a graphical user interface for model
specification based on the MATLAB/SimuLink framework.
Keywords: Real time analysis, network calculus, scheduling, cyclic flows, timed automata, tools
