Space-based multi-core programming in Java

Abstract

Multi-core processors require programmers to exploit concurrency in software as far as possible. Unfortunately, our current concurrency abstractions make multi-core programming harder than necessary because we have to reduce unintended non-determinism on a very low level of abstraction, for instance via synchronisation mechanisms. In this experience paper we analyse if space-based systems can mitigate multi-core programming in the Java programming language and present the Procol programming model that introduces the space-based choreography of active components, which internally orchestrate fine-grained workflow activities. The main contributions are (1) the Procol programming model, (2) benchmark results of the scalability of different tuple space implementation techniques that we evaluated on different multi-core architectures, (3) benchmark results of the scalability of two equivalent Mandelbrot applications for best-case measurements -- one implemented with the standard Java thread model, one with our Procol programming model. The conclusions drawn from these experiments are (1) tuple space data structures that scale reasonably well can be provided, (2) the performance overhead that Procol imposes is at least for the considered application a reasonable trade-off for the ease of programming provided for multi-core architectures.