Faster Sorting for Flash Memory Embedded Devices

Abstract

Embedded devices collect and process data in a wide variety of applications including consumer and personal electronics, healthcare, environmental sensors, and Internet of Things (IoT) deployments. Processing data on the device rather than sending it over the network for analysis is often faster, more energy efficient, and supports decision-making closer to data collection. A fundamental data manipulation operation is sorting. Sorting on embedded devices with flash memory is especially challenging due to the very low memory and CPU resources. Previous work developed customized algorithms that avoided writes and minimized memory usage. The standard external merge sort algorithm has limited application on small devices as it requires a minimum of three memory buffers and is not flash-aware. The contribution of this work is an extension of external merge sort that requires only two memory buffers and is optimized for flash memory. The result is an algorithm that improves on the state-of-the-art and applies to a wider range of devices. Experimental results demonstrate that when sorting large data sets with small memory the algorithm reduces I/Os and execution time by about 30%.