Edge Workload Prediction Based on Deep Learning

In the post-COVID-19 era, the rapid growth of edge computing underscores the importance of accurate workload prediction to maximize utilization of limited edge resources. Both edge service providers (ESPs) and edge service consumers (ESCs) can benefit significantly from such predictions. Existing workload prediction paradigms (edge-only or cloud-only) fall short by neglecting inter-site correlations and encountering data transmission delays. Given the expanding adoption of edge platforms by web services, achieving accuracy and efficiency in workload prediction is of vital importance. This chapter introduces a cloud-edge collaborated edge workload prediction framework that employs a collaborative cloud-edge paradigm for edge workload prediction using multi-view graphs. Specifically, the global stage develops a learnable aggregation layer per edge site to enhance efficiency while capturing inter-site correlations. The local stage devises a disaggregation layer that merges intra-site and inter-site correlations to enhance prediction accuracy. Through extensive experimentation on real-world edge workload data from China's largest edge service provider, the proposed workload prediction framework outperforms existing methods in minimizing time consumption and reducing communication costs.