Sibyl: Host load prediction with an efficient deep learning model in cloud computing

Abstract

Prediction of host load is essential in Cloud computing for improving resource utilization and achieving service-level agreements. However, accurate prediction of host load remains a challenge in Clouds because the type of load varies differently. Furthermore, selecting metrics for host load prediction is also a difficult task. With so many metrics in the Cloud systems, it is hard to determine which metrics are going to be useful. To address these challenges, this paper proposes an efficient deep learning model named Sibyl to improve the accuracy and efficiency of prediction. Sibyl includes two parts: a metrics selection module and a neural network training module. Sibyl first selects metrics by filtering out irrelevant metrics. Afterwards, Sibyl applies a powerful neural network model built with bidirectional long short-term memory to predict actual load one-step-ahead. We use Sibyl to analyze a 40-day load trace from a data center with 176 machines. Experiments show that Sibyl can reduce training metrics while maintaining prediction accuracy. Besides, Sibyl significantly improves prediction accuracy compared to other state-of-the-art methods based on autoregressive integrated moving-average and long short-term memory.