Asia-Pacific Network Operations and Management Symposium
IT Resource Trend Analysis by Component Decomposition Based on Non-negative Matrix Factorization
Yuji SAITOH, Tetsuya UCHIUMI, Yukihiro WATANABE,
PDF download (438.1KB)
In the field of ICT systems, understanding the long-term trends of resource usage is useful for efficient resource provisioning. Additionally, it is more useful to understand the trends for each business process running on a device. However, generally, analysis for each business process is not possible since the acquired resource data contains the resource data of multiple business processes. In this paper, we propose a novel analysis method that decomposes resource data into components of each business process. The proposed method is based on non-negative matrix factorization (NMF), which is generally applied for decomposing a sound signal into signals of musical instruments and vocals. By NMF, the resource data acquired for each device is decomposed into multiple components with different business characteristics, and the long-term trend for each component is estimated. As a result, it is possible to detect increases in the long-term load of a specific business process, which is not possible with conventional approaches. Finally, we evaluate the proposed method by using simulated data and real data and show that the proposed method can decompose resource data into components of each business process.