A Novel LSTM Network For Time Series Forecasting Of Network Traffic In Cloud Computing Environments
Open Access DepositedThe forecasting of real time network traffic accurately and efficiently is an ongoing and complex challenge in cloud computing applications. It is not only essential for efficient resource allocation but also for cost optimization, and maintaining Quality of Service (QoS). However, the dynamic and unpredictable nature of workloads often creates complexity in traditional forecasting methods. This praxis demonstrates a Cloud Network Traffic Optimization Model (CNTOM) which is a novel deep learning architecture designed to address this challenge. This model, CNTOM, enhances a Long Short-Term Memory (LSTM) network with a self-attention mechanism, thereby enabling it to dynamically weigh the importance of historical data points when making predictions. The dataset used is a large-scale, real-world cluster data trace for machine metrics from Alibaba cloud, we systematically evaluated the performance of CNTOM against a suite of baseline models, including Vector Autoregression (VAR), LightGBM, Multilayer Perceptron (MLP), and standard LSTM variants. One of the key aspects of this research is analyzing the impact of applying a Savitzky-Golay smoothing filter as a data preprocessing step. Doing this, created parallel smoothed and unsmoothed data streams for a direct comparative analysis. A Normalized Root Mean Squared Error (NRMSE) was used to measure performance and to ensure a fair, scale-independent comparison. The praxis demonstrates that the proposed CNTOM model, which was trained on the smoothed dataset, achieved superior performance, yielding the lowest NRMSE of 0.1356. This showcases a substantial improvement of nearly 48% over the unsmoothed VAR baseline, demonstrating a clear advantage over all other architectures that were tested. The study also clearly confirmed that data smoothing is a crucial preprocessing step. Data smoothing dramatically improved the accuracy of all models. The NRMSE reductions ranged from 27% for the VAR model to over 41% for the CNTOM model.The findings of the praxis highlight the combined benefits of a sequence-aware architecture, an attention-based focus mechanism, and needed data preprocessing. Through this research, we conclude that the attention-enhanced CNTOM architecture, when combined with signal smoothing, delivers a powerful and highly effective approach for predicting machine-level network traffic in dynamic cloud environments.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.