Adaptive checkpoint placement in energy harvesting real-time systems

Most of the current embedded systems work in unsecure and fault-prone environments. Such systems usually harvest their required energy from the environment with many sources of unpredictability. On the other hand, more than often, these systems have timing constraints that should be satisfied in spite of the existing problems and limitations mentioned above. In this paper, we propose two checkpoint placement algorithms to manage the harvesting energy through dynamic voltage scaling (DVS) while satisfying the system timing constraints. An offline algorithm is proposed that unformly distributes the checkpoints according to the predicted harvesting energy from the environment. An online algorithm is also proposed which takes the harvesting energy variability into account and dynamically adjusts the checkpoint placement as well as the processor voltage and speed. Therefore, the latter algorithm non-uniformly distributes the checkpoints. The obtained reward by these algorithms have been compared in different energy and workload scenarios through simulation. According to the comparisons, different situation that each of these algorithms may be preferrable are specified.