Since multi-processor systems have benefits of scalability in performance and power, they become prevalent in embedded systems as well as desktop computers. A critical issue of designing a multiprocessor embedded system is to evaluate the expected per...
Since multi-processor systems have benefits of scalability in performance and power, they become prevalent in embedded systems as well as desktop computers. A critical issue of designing a multiprocessor embedded system is to evaluate the expected performance early in the design process before actual implementation. For the application domains that have hard real-time constraints, it is essential to estimate the worst-case performance of the application. In particular we are interested in the worst-case response time (WCRT) of an application once it is triggered. In an automotive system, for example, an application algorithm is triggered by the arrival of a sensor input and the computation result should be delivered to actuators on time. However, it is a very challenging problem to determine a tight upper bound of the worst-case response time in a distributed real-time embedded system when execution time variations of tasks, jitter of input arrivals, scheduling anomaly behavior in a multi-tasking system, and heterogeneity of processing elements, are considered all together.
In the first part of this dissertation, we propose a novel analytical method, called STBA (Scheduling Time Bound Analysis) technique, which has a polynomial time complexity to the number of tasks while producing a much tighter bound than well-known compositional approaches, SymTA/S and MPA. Unlike conventional schedulability analysis approaches that are based on preemptive scheduling policy, the proposed technique supports an arbitrary mixture of preemptive and non-preemptive processing elements. In the STBA technique, by analyzing the graph topology and worst-case scheduling scenarios on the mapped processing element, we measure the schedulable time bound of each task as tightly as possible, which is characterized by two pairs of timing information: (minimum start time, maximum start time) and (minimum finish time and maximum finish time)
In the second part of dissertation, we propose a novel optimal method based on MILP (Mixed Integer Linear Programming) that is a well-known technique for combinatorial optimization problems. We translate the worst case response time problem as an optimization problem to use MILP solver. In the proposed technique, we formulate a set of MILP formula in a compositional way for modeling flexibility, but solve the problem holistically to achieve exact upper bounds. To mitigate the time complexity of the MILP method, we exploit the result from STBA technique to reduce the number of variables and confine the variable ranges.
As another method, we propose a meta-heuristic based method for the performance analysis. Like an optimal method we translate WCRT problem to optimization problem, but use meta-heuristic algorithm to solve the problem. In this method, we cannot achieve exact worst case response time. However, it overcomes the main limitation of optimal method that is exponential time complexity. Furthermore, we can get many near worst case scenarios which are close to exact worst scenario as population converged.
Extensive experiments are performed to confirm that the proposed technique gives reasonable performance for analysis. A real-life example shows promising results that give tight bounds in an affordable solution time.