Software patterns for asymmetric multiprocessing devices on embedded systems: a performance assessment

In embedded systems there is a variant of Multicore System on Chip devices (MSoC devices) where not all the computing elements (processor cores) are equal. The differences in the cores of these devices range from different hardware architectures using the same instruction set to completely different processors working together inside the same device. These SoCs are called “Asymmetric Multi Processing Devices” (AMP Devices). In order to help developers to take advantage of the possibilities that these devices may offer in the context of embedded systems, software design patterns have been defined, describing software architectural solutions with known uses. However, there are still no experimental results showing the benefits of these solutions. In this work we measure the performance of a design pattern called Mini Me, applied on an AMP device configuration, and compare it against two Symmetric Multiprocessing Device (SMP Device) configurations. The evaluations show a better than expected computing performance of the AMP Configuration using the design pattern Mini Me.