INTELLIGENT WORKLOAD SCHEDULING USING HYBRID DEEP LEARNING IN MULTI-CLOUD PLATFORMS

• Sadeq Thamer Hlama

  Department of Computer Science College of Science, University of Sumer, Iraq, Dhi-Qar, Iraq

  Author

Cloud computing services have proliferated over the years and a lot of organizations have started to adopt a multi-cloud architecture crossing different cloud providers to provide better workload availability, scalability, and performance. However, considering dynamic resource requirements, heterogeneous setup and variable network conditions efficient scheduling of workloads continue to be a major challenge. Conventional scheduling policies do not usually generalize well to more complex and dynamic environments. Towards this end, we present a hybrid deep learning based intelligent workload scheduling framework for on-demand computing in multi-cloud platforms. We perform a combination of feature extraction through Convolutional Neural Networks (CNN) and temporal workload prediction using a Bidirectional long short-term memory (BiLSTM) network. It enables dynamic workload management through predicting the network and distribution of tasks over multiple clouds to enhance both system performance as well as resource utilization. Experimental results show that our proposed task scheduling method outperforms various traditional and state-of-the-art scheduling methods including round robin, genetic algorithm, and standard LSTM-based schedulers in terms of load balancing efficiency, task completion time, and resource utilization. These findings are of great significance and reflect the effectiveness of hybrid deep learning methods in achieving intelligent resource management in a distributed cloud environment.

• pdf

This work is licensed under a Creative Commons Attribution 4.0 International License.