ARM Server Error Prediction based on Fusion Features
Chenyue Zhang, Shipeng Ji, Mingqing Zhang, Zhixiang Jiang · 2024
Aiming at the problem that the ARM server has weak error handling capability due to the lack of efficient error prediction mechanism, this paper deeply explores the features of ARM server logs and memory correctable errors, and proposes a fusion prediction method LMF that can make full use of ARM server log features and memory correctable error features. In order to improve the running efficiency of the model, a strategy based on threshold and cycle to update the log template is proposed, and the inference time is introduced in the evaluation of the LMF model. Experiments on the labeled dataset show that compared with the use of log predictors alone, the fusion method has improved both precision and recall. The relevant indicators are tested and verified on the actual ARM server cluster, which shows that the LMF method can improve the error detection and handling capabilities of the server.