It has been challenging to predict data in terms of monitoring information technology (IT) resources. In order to obtain the quality and performance of products, changes can be detected and monitored setting up a fixed threshold value based on statistics and operation experiences. Monitoring data by a fixed threshold value may not work properly during none busy hours in exceptional situations whereas a usage change during busy hours can be detected. It is because it cannot reflect the trend of resource usage seasonality as a function of time. The technique based on Time Series Decomposition (TSD) can provide the one with appropriate methodology so that problems can be recognized and diagnosed and the correction can be made ahead of time by detecting a subtle status change of devices in massive IT resources. In this paper, we propose three approaches to predict data such as Intelligent Threshold, Abnormal Pattern Detection, time prediction of reaching target value; the appropriate trend detection of Time Series, optimal seasonality detection and technique using Log Regression Seasonality. The experimental data collected here exhibit that it can reflect the change over time to the prediction data improving its accuracy compared to existing TSD technique.
展开▼
