A novel approach to workload prediction using attention-based LSTM encoder-decoder network in cloud environment

• DOI: 10.1186/s13638-019-1605-z
• Publication Date: 2019-12-17T19:03:02Z
• Authors: Yonghua Zhu; Weilin Zhang; Yihai Chen; Honghao Gao
• Keywords: TK7800-8360; Telecommunication; 0202 electrical engineering, electronic engineering, information engineering; Workload prediction; Attention mechanism; TK5101-6720; 02 engineering and technology; Electronics; LSTM; Encoder-Decoder Network; Cloud environment

Abstract Server workload in the form of cloud-end clusters is a key factor server maintenance and task scheduling. How to balance optimize hardware resources computation should thus receive more attention. However, we have observed that disordered execution running application batching seriously cuts down efficiency server. To improve prediction accuracy, this paper proposes an approach using long short-term memory (LSTM) encoder-decoder network with attention mechanism. First, extracts sequential contextual features historical data through encoder network. Second, model integrates mechanism into decoder network, which for batch workloads can be carried out. Third, experiments out on Alibaba Dinda traces dataset demonstrate our method achieves state-of-the-art performance mixed cloud computing environment. Furthermore, also propose scroll method, splits sequence several small sequences monitor control accuracy. This work helps dynamically guide configuration balancing.