Processor and Bus Co-scheduling Strategies for Real-time Tasks with Multiple Service-levels

Cyber-Physical Systems, including those in the automotive domain, are often designed by assigning to each task an appropriate criticality-based reward value which is acquired by the system on its successful execution. Additionally, each task may have multiple implementations designated as service-levels, with higher service-levels producing more accurate results and contributing to higher rewards for the system. This work proposes strategies for co-scheduling a set of periodic tasks with multiple service-levels, on homogeneous processors and system buses. The problem is modeled as a Multi-dimensional Multiple-Choice Knapsack formulation (MMCKP) with the objective of maximizing overall system level rewards. A Dynamic Programming (DP) solution is proposed to solve the MMCKP. It was observed that although the DP based solution produces optimal results, its complexity is highly sensitive to the number of tasks, processors, buses as well as to the number of task service-levels, which severely restricts scalability of the strategy. Therefore, we have also proposed a fast yet efficient heuristic algorithm called Accurate Low Overhead Level Allocator (ALOLA), which attempts to achieve the same objective. Our simulation based experimental evaluation shows that even on moderately large systems consisting of 90 tasks with 5 service-levels each, 16 processors and 4 buses, while MMCKP incurs a run-time of more than 1 hour 20 minutes and approximately 68 GB main memory, ALOLA takes only about 196 $\mu s$ (speedup of the order of 106 times) and less than 1 MB of memory. Moreover, while being fast, ALOLA is also efficient being able to control performance degradations to at most 13% compared to the optimal results produced by MMCKP. We use an automated flight control system employed in modern avionic systems, a real-world application to illustrate the general applicability of our proposed scheme.