An asynchronous nonblocking coordination and synchronization protocol for a parallel robotic control kernel

Abstract

Over the last 25 years, performance improvements by the steady increase of CPU clock frequencies were the driving factor for innovations in the domain of computationally intensive embedded applications. Now the free lunch is over [12] -developers have to parallelize their systems in order to achieve further improvements by integration of multi-core platforms. In embedded systems, this is even more challenging than in the domain of desktop computers, as safety properties and hard real-time constraints impose a much stronger demand on determinism. In this experience report, we present a concrete coordination and synchronization problem for a double buffering procedure that arose on our ongoing attempts to parallelize a robotic control kernel. This double buffering procedure used by two tasks must assure a consistent data flow without data losses. Therefore, we approach a fast bounded wait-free solution, which does not suffer from priority inversion.