Online Prediction of Server Crash Based on Running Data
Zhuoliang Zou, Jun Ai · 2020
For web servers, the most typical and common failure is that the client's web page requests surge in a certain period of time, resulting in the server's collapse under extreme pressure. The early warning of server crash time provides the possibility to avoid crash loss. In this paper, we propose an early-warning method for Web service failure. Firstly, we obtain the key nodes on the running path of the server, collect and analyze the running status data of the path under various pressures through program instrumentation, and inform the occurrence of failure in advance through LSTM-SVM (the algorithm combining SVM and LSTM). We apply this method to Nginx, a widely used server, and the accuracy of crash warning is over 95%. Experiments show that the method of acquiring target data has little effect on the performance of server, achieves high degree of automation, and realizes high-precision fault warning. The innovation of this paper is that we realize the fault warning through the change of the parameters in the software.