Optimization of big data transfers among data center networks using deep reinforcement learning with graph neural networks

The big data era is an emerging paradigm that has gained a lot of interest in the last few years from industry, academia, and governments around the world. Cloud computing infrastructure often operates over multiple distributed data centers around the globe, following a pay-as-you-go pricing model. Enabling fast data transfer across these data centers, with low monetary cost and without link congestion, is not a trivial task. Efficient protocols and tools are necessary to transfer a huge amount of data while taking into account the user’s quality of service (QoS) requirements. With the recent widespread use of artificial intelligence (AI) and its application in network optimization scenarios, deep reinforcement learning (DRL), which combines reinforcement learning with deep learning, has emerged as a prominent approach for big data transfers among data center networks. In this article, we introduce a novel approach that integrates DRL with graph neural networks (GNN) to come up with an efficient strategy for big data transfer. Our approach generates continuous control actions to optimize data transfer. It can learn from past actions and successfully generalize to different incoming scenarios. Results show that our method consistently optimizes big data centers among data centers.