Towards complex dynamic fog network orchestration using embedded neural switch. Issue 2 (7th February 2021)

• Record Type: Journal Article
• Authors: Okafor, K. C.; Ononiwu, G. C.; Goundar, Sam; Chijindu, V. C.; Udeze, C. C.
• Is Part Of: International journal of computers and applications. Volume 43:Issue 2(2021)
• Page Start: 91
• Page End: 108
• Publication Date: 2021-02-07
• Subjects: Fog computing -- cloud network -- artificial intelligence -- QoS complexity -- network congestion -- big data; Computers -- Periodicals; Computer software -- Periodicals; Computer networks -- Periodicals; Multimedia systems -- Periodicals; Internet -- Periodicals; World Wide Web -- Periodicals; Minicomputers -- Periodicals; Microcomputers -- Periodicals; 004.05
• DOI: 10.1080/1206212X.2018.1517440
• Languages: English
• ISSNs: 1206-212X

Abstract

ABSTRACT: Cloud data centers used for High Performance Computing (HPC) with volatile Internet of Things (IoT) devices absolutely requires zero-speed switching/low latency, normalized throughput, network stability with near zero packet drops during heavy traffic workload. Deadlock traffic condition found in baseline Fog-nodes results when there is no dynamic provisioning of services at Fog layer. In this case, Quality of Service (QoS) metrics of Service Level Agreements (SLA) are violated. Motivated by these concerns, this paper proposes Smart Hierarchical Network (SHN) as a reliable Fog dynamic design structure based on Software Defined Artificial Neural Network (SD-ANN). It features congestion-aware neural switch model with embedded predictive receding horizon for intelligent congestion management. The goal is to exploit the locally optimized Fog neural switch connections and maximize the overall QoS while satisfying the enormous traffic workload requirements. A sampled real-world trace-file workload from Galaxy backbone, Nigeria, is compared with SHN for Fog service provisioning. It is shown that with receding horizon, the ANN-based model ideally offers 100% throughput R value. Under the established training scenarios, the ANN switch offers the lowest mean square error while yielding acceptable QoS metrics. The result is significant for scalable networks supporting massive computational workloads.