A Timed-Automata Approach for Critical Path Detection in a Soft Real-Time Application

In this paper, we report preliminary ideas from our project called " Time Performance Improvement With Parallel Processing Systems " (TIPS). In the TIPS project, we plan to take advantage of multi-core platforms for performance improvement by parallelizing a complex soft real-time application implemented in Java. Manual parallelization does not generally yield optimal performance since its results depend on the designer's intuition and experience; it is not feasible to manually inspect all possible parallelization alternatives and their impact on the performance for large-scale and complex software systems. Even if we are able to find possible parallelization alternatives with tool support, not all parallelization efforts end up with a performance improvement. One needs to detect the execution paths of the application which are semantically relevant and have higher timing costs. By the term semantically relevant, we mean the execution paths that play an important role from the perspective of requirements of the system. These paths are called " critical paths " and an improvement in these paths is likely to lead to better overall performance of the application whereas the improvements for non-critical paths might not guarantee this [3][4][5][6][7][8]. By examining the critical paths, we can find the bottlenecks along them [3]. The problem we are tackling is that: How can we find the critical execution paths of the application that violate the timing constraints? As sub-problems, we need to answer the following questions: (i) how can we model the application formally so that we can search for critical paths effectively and (ii) how can we represent critical path definitions so that these definitions can give us critical paths when they are applied to the model. One can choose to apply source code analysis directly to get a formal model of the application. However, this approach has two shortcomings. First, source code analysis does not scale for such a large system as we have in the TIPS project. And second, one cannot derive the relevance of an execution path just by observing the source code; relevance has to be derived from the requirements to decide how relevant a critical path is from the requirement analysis perspective. The approach we propose for critical path detection is shown in Fig. 1. It is based on the creation and usage of a formal timed-automata model of the application which contains the information related to the execution steps with timing properties. Then, we query this …