Deep Parameter Tuning of Concurrent Divide and Conquer Algorithms in Akka

Abstract

Abstract. Akka is a widely-used high-performance and distributed computing toolkit for fine-grained concurrency, written in Scala for the Java Virtual Machine. Although Akka elegantly simplifies the process of building complex parallel software, many crucial decisions that affect system performance are deferred to the user. Employing the method of Deep Parameter Tuning to extract embedded 'magic numbers' from source code, we use the CMA-ES evolutionary computation algorithm to optimise the concurrent implementation of three widely-used divide-and-conquer algorithms within the Akka toolkit: Quicksort, Strassen's matrix multiplication, and the Fast Fourier Transform.