SHIELD: A secure heuristic integrated environment for load distribution in rural-AI

Kaushal, Ashish, Almurshed, Osama, Almoghamis, Osama, Alabbas, Areej, Auluck, Nitin, Veeravalli, Bharadwaj and Rana, Omer ORCID: https://orcid.org/0000-0003-3597-2646 2024. SHIELD: A secure heuristic integrated environment for load distribution in rural-AI. Future Generation Computer Systems: The International Journal of eScience 161 , pp. 286-301. 10.1016/j.future.2024.07.026 Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 17 July 2025 due to copyright restrictions. Download (2MB)

Abstract

The increasing adoption of edge computing in rural areas is leading to a substantial rise in data generation, necessitating the need for development of advanced load balancing algorithms. This is particularly important in applications that utilise existing, though limited, computational and data communication infrastructures. Furthermore, rural communities have growing concerns regarding the privacy, security, and ownership of the data produced within their agricultural fields. Load distribution in rural edge devices can enhance agricultural practices by improving resource usage, decision-making, and addressing network connectivity challenges. Managing resource utilisation in this way also improves economic investments made in managing and deploying edge devices in rural environments. In this work, we propose SHIELD, a security-aware load balancing framework, primarily designed for edge-based systems in rural areas. For handling environments with limited connectivity, SHIELD efficiently manages tasks and computational resources by categorising them into restricted, public and private, shared respectively. It also allocates tasks considering key performance factors such as completion time, resource utilisation, failure rate, and security. The framework is evaluated on a weed detection scenario in precision agriculture, using three federated learning (FL) variants (local model training, global model aggregation, and model prediction) with the ResNet-50 model trained on the DeepWeeds image classification dataset. The proposed framework also integrates encryption and task replication techniques for data confidentiality, integrity, and availability. Experimental results show that SHIELD demonstrates an average of 23% (using Parsl), 29% (using OpenWhisk) improvement in failure rate and 18s (Parsl), 13s (OpenWhisk) average improvement in makespan compared to other task allocation approaches, such as secure variants of random, round robin, and least loaded.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Computer Science & Informatics
Additional Information:	License information from Publisher: LICENSE 1: Title: This article is under embargo with an end date yet to be finalised.
Publisher:	Elsevier
ISSN:	0167-739X
Date of First Compliant Deposit:	29 July 2024
Date of Acceptance:	13 July 2024
Last Modified:	30 Jul 2024 07:18
URI:	https://orca.cardiff.ac.uk/id/eprint/170821

Actions (repository staff only)

Edit Item