Storage Performance Prediction

Abstract

Performance is one of the most important capabilities provided by storage systems. How to accurately understand the performance of storage systems and predict future performance requirements for storage is a major challenge. The traditional method is to avoid the problem through Over Provisioning. Buying more and more advanced storage devices. This solution greatly increases the purchase cost of data center. In this paper, dynamic time wrapping is introduced on the basis of fbprophet algorithm. The accuracy of storage performance prediction is further improved. Experiments show that our method can reduce the deviation value of MAPE to less than 10%. It can effectively provide the basis for storage managers to expand capacity and upgrade system.