Accurate workload prediction for edge data centers: Savitzky-Golay filter, CNN and BiLSTM with attention mechanism

Workload prediction is a fundamental task in edge data centers, which aims to accurately estimate the workload to achieve an in-situ resource provisioning for workload execution. In this paper, we propose a deep learning model termed SG-CBA to predict workload, which is powered by Savitzky-Golay filter (SG filter), Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) with Attention mechanism. First, raw time series of the workload is normalized and smoothed by a preprocessing module with SG filter. Following that, we establish a deep learning module based on CNN and BiLSTM with attention mechanism to extract and process the features for the accurate workload prediction. Real-world workload from Alibaba cluster is adopted to validate our proposed model in the experiments. Experimental results demonstrate that SG-CBA can achieve accurate workload prediction, which outperforms the alternatives, including BTH-ARIMA, LSTNet, OCRO-MLNN, Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), LSTM and BiLSTM under different evaluation metrics.

Automated summary

In this paper, a deep learning model SG-CBA is proposed for workload prediction, which combines SG filter, CNN, and BiLSTM with attention mechanism. The experimental results demonstrate that SG-CBA outperforms other alternatives in accurately predicting workload under various evaluation metrics.