Foundations for sustainable and trustworthy edge data analytics

Abstract

Massive amounts of data are continuously generated from a growing number of Internet of Things (IoT) devices. Based on the insights obtained through the analysis of collected data, different data-driven decisions are made to manage IoT systems. Traditionally, managing such systems includes data processing in the cloud. However, performing data processing in centralized cloud data centers brings serious challenges, including the transfer of huge amounts of sensor data over the network and new strict requirements (e.g., latency, accuracy, privacy) from IoT applications (e.g., smart buildings, smart traffic). For these reasons, edge computing has been introduced. Edge computing represents a promising methodology and solution to execute analytics close to data sources using much smaller edge servers and devices. However, scalable and centralized cloud services cannot be generalized and directly applied to edge infrastructures. Also, IoT decision-making processes require timely and accurate data processing, bringing a new set of challenges to design sustainable and trustworthy edge data analytics. This is because performing edge data processing needs to deal with problems such as limited computational and storage resources; sensor data that can often be incomplete leading to inaccurate analytics; decentralized data locations posing difficulties for latency-critical analytics placement. We address these problems by targeting data-centric perspectives for sustainable and trustworthy edge analytics. We introduce an adaptive data recovery mechanism for incomplete sensor data, improving the accuracy of data analytics and decision-making processes. Further, we propose an efficient edge storage management mechanism for keeping only the most relevant data in limited edge storage. We also propose a self-adaptive and data locality-aware edge analytics placement mechanism that minimizes the latency for performing edge analytics. We show the integration of edge computing and modern network communication technologies in a prototype deployment, ensuring reliability and privacy for critical IoT applications. Finally, we evaluate the proposed solutions using real-world datasets, applications, self-designed benchmarks and testbeds using physical edge infrastructures. Besides novel edge data services and approaches making the theoretical contribution, we also show their practical applicability in IoT systems. The proposed solutions can help IoT systems to produce more accurate, reliable, and timely decisions, thus, contributing to foundations for sustainable and trustworthy edge data analytics.