Optimized Allocation of Fault-tolerant Embedded Software with End-to-end Timing Constraints

It is desirable to optimize power consumption of distributed safety-critical software that realize fault tolerance and maximize reliability as a result, to support the increasing complexity of software functionality in safety-critical embedded systems. Likewise, safety-critical applications that are required to meet endto-end timing constraints may require additional computing resources. In this paper, we propose a scalable software-to-hardware allocation based on hybrid particle-swarm optimization with hill-climbing and differential algorithms to efficiently map software components to a network of heterogeneous computing nodes while meeting the timing and reliability constraints. The approach assumes fixed-priority preemptive scheduling, and delay analysis that value freshness of data, which is typical in control software applications. Our proposed solution is evaluated on a range of software applications, which are synthesized from a real-world automotive AUTOSAR benchmark. The evaluation makes comparative analysis of the different algorithms, and a solution based on integer-linear programming, which is an exact method. The results show that the hybrid with the hill-climbing algorithms return very close solutions to the exact method and outperformed the hybrid with the differential algorithm, though consumes more time. The hybrid with the stochastic hillclimbing algorithm scales better and its optimality can be deemed acceptable.