A Speculative Computing Approach to Accelerate the Task Execution in Cyber-Physical Systems

The manufacturing world is increasingly empowered by concepts like the Internet of Things (IoT), data analytics or speculative computing. Speculative computing lies in two ideas: i) predicting the output of specific tasks to avoid executing them, ii) computing specific heavy tasks in idle times so that, when their output is needed, other tasks do not have to wait for those heavy tasks to execute. This paper presents a solution to decrease the execution time of task pipelines in distributed Cyber-Physical Systems (CPS) with the help of speculative computing. The system learns how to map a task input into its output and then predicts new inputs' results using lookup tables and speculators. Speculators learn continuously and predict with a particular error, indicating the confidence of the speculated value. The work was evaluated using as testing scenario the Travelling Salesman Problem (TSP) based on genetic algorithms.