TAFLE: Task-Aware Flow Scheduling in Spine-Leaf Network via Hierarchical Auto-Associative Polynomial Reg Net

Abstract

Cloud computing has become crucial to modern infrastructure, which enables data-intensive applications to thrive in scalable environments. The backbone of cloud computing is the massive data center (DC) servers. The DC networks have unique traffic demands for different tasks, which need to be considered for efficient network traffic (NT) management and enhancing Quality of Service (QoS). Existing solutions fail to consider these unique traffic demands, which result in suboptimal performance in large-scale, data-sensitive environments. To overcome these challenges, a novel Traffic-aware FLow reconfiguration in spine lEaf (TAFLE) system has been proposed in this paper. The proposed model addresses the inefficiencies of QoS-based network traffic allocation by considering the task-level requirements of data-sensitive applications. The proposed solution combines a Deep Packet Analytics (DPA) engine and the Hierarchical Auto-Associative Polynomial Reg Net (HAP-Reg Net) model for reconfiguring the flow based on QoS classes and predicted traffic volumes. Several criteria have been used to evaluate the proposed TAFLE model, such as the f1-score, accuracy, precision, recall, Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). Experimental findings show that the system significantly improves prediction accuracy and resource allocation, which leads to better overall performance. Experimental results demonstrate that the existing techniques, such as CNN, LSTM, and GRU models, achieve 96.12%, 96.08%, and 95.65% accuracy, while the novel HAP-Reg Net model achieves 96.49% accuracy. Additionally, the proposed TAFLE model has a greater accuracy of 99.2% than previous methods like VAMBIG, AMFQ, and D-LSLP, which have 95.51%, 97.35%, and 98.89% accuracy, respectively.