New Framework Boosts Cloud-Edge Workload Forecasting Accuracy

A new research paper introduces DSTFView, a sophisticated framework aimed at improving workload forecasting in complex cloud-edge computing environments. With the increasing reliance on edge AI for latency-sensitive applications, accurate prediction of resource demands is critical. Existing forecasting methods often struggle to balance the modeling of diverse features with computational efficiency in these distributed settings. DSTFView addresses these challenges by employing a dual-input spatio-temporal-frequency modeling approach. This allows it to simultaneously capture closeness and period dependencies, as well as extract intricate patterns across spatial, temporal, and frequency domains. A key innovation is its adaptive fusion mechanism, which dynamically adjusts the contribution of each data view to effectively respond to sudden shifts in workload. Experimental evaluations conducted on CPU and TP datasets demonstrate that DSTFView consistently surpasses the performance of current state-of-the-art forecasting methods. It shows superior accuracy across various forecasting horizons and evaluation metrics, making it a valuable tool for optimizing resource allocation and ensuring reliability in cloud-edge infrastructures.