An Autonomous Task-Clustering Manager for Edge Computing Applications Running on Raspberry Pi
An Autonomous Task-Clustering Manager for Edge Computing Applications Running on Raspberry Pi

Summary/Abstract: Edge nodes in Edge computing systems are often deployed in close proximity to IoT devices for local data processing and storage. Recently, it has been shown that low-end single-board computers, such as Raspberry Pis, can be utilized as Edge nodes. However, these devices are limited in terms of computational capabilities, especially in cases where they need to perform a large number of concurrent tasks in response to connected IoT devices. Furthermore, these edge nodes may be deployed in unstable or hostile environments with fluctuating power sources, affecting their performance. This paper investigates how task clustering techniques can achieve different configurations of task clusters running on a Raspberry Pi as an Edge node. First, an empirical evaluation is conducted to examine how various task-cluster configurations affect the performance of applications running on these devices. Experimental results show that up to 25% improvement in execution time is obtained using a configuration with an appropriate task clustering configuration compared to a configuration in which each object is scheduled as a separate task. Then, this study describes the design of an autonomous Task-Clustering manager extension based on the MAPE-K feedback loop that runs as a layer on top of existing job schedulers to automatically configure task clusters running on these devices to improve performance.

• Details
• Contents