Memory Error Driven Server Failure Detection
Rafael Mozart da Silva, Francisco Lucas F. Pereira, Victor A. E. Farias, Felipe T. Brito, Javam C. Machado · 2024
The correct functioning of Dynamic Random Access Memory (DRAM) is of fundamental relevance to the functioning of servers in data centers. Therefore, being able to detect server failure caused by memory errors is fundamental to the development of prediction methods that can be used to avoid server failure caused by memory errors. Thus, ensuring the continuous availability of the hosted services. In recent years, many authors proposed machine learning-based methods to predict server failure based on the occurrence of DRAM errors. However, from previous works, one can notice that this is a challenging task due to the lack of data and the irregularity in which memory errors occur. In this work, through feature engineering, we look forward to improving the classification accuracy of recurrent neural networks at dealing with irregularly sampled data in order to improve the accuracy in identifying servers that are nearing a failure state.