Algorithms for Optimally Tracing Time Critical Programs

The current work is the closing article of an ongoing series of papers that illustrate a high-performance solution to the tracing of time critical applications with minimal impact on the running system. The idea of the proposed solution is to instrument the application at significant points with event functions, log the events of these functions during execution and reconstruct the complete control flow on the basis of protocoled events. The finding of the minimum set of the monitoring points in the control flow graph, which is sufficient to reconstruct all posible control flows, is a NP-complete problem (FeedbackVertex-Set (FVS) [Gar79]). In this paper we present the graph-theoretical algorithms for the selection of the monitoring points in the control flow graphs, which are based on decomposition of graphs into connected components. We present the solution, which combines the graphtheoretical algorithms with the Knuth-Stevenson Algorithm and provides in most cases the minimal set of monitoring points. But generally this problem is still NP-complete. This paper deals with results from a joint research project between the Siemens AG, COM MN CC ST, Berlin and the University of Oldenburg.