Time-Dependent Cloud Workload Forecasting via Multi-Task Learning

Abstract

Cloud services have rapidly grown in cloud data centers (CDCs). Accurate workload prediction benefits CDCs since appropriate resource provisioning can be performed for their providers to ensure the full satisfaction of service-level agreement (SLA) requirements from users. Yet these providers face some challenging issues in accurate workload prediction, i.e., how to achieve high accuracy and fast learning of prediction models. Consistent efforts have been made to address them. This letter proposes an innovative integrated forecasting method that combines stochastic configuration networks with Savitzky–Golay smoothing filter and wavelet decomposition to forecast workload at the succeeding time slot. We first smooth the workload via a Savitzky–Golay filter. Then, we adopt wavelet decomposition to decompose smoothed outcome into multiple components. Supported by stochastic configuration networks, an integrated model is established, which can well describe statistical features both of detail and trend components. Extensive experimental outcomes have explicated that our approach realizes better prediction results and quicker training than those of representative prediction approaches.