Why is it so hard to predict computer systems failures?
Jomar Domingos, Raul Barbosa, Henrique Madeira · 2021
Proactive fault management represents a promising path that could significantly enhance computer systems dependability, allowing to deal with failures before their occurrence. Online failure prediction plays a very important role in proactive fault management schema, being the first stage of the process. However, in spite of decades of research on the topic, and the very high relevance of failure prediction, the reality is that we still do not have readily available failure prediction methods and guidelines. This paper presents a reflection on the most relevant challenges faced by failure prediction in computer systems and discusses why it has been so difficult to develop effective methods and tools that can be relied upon to predict failures in real computer systems. Along with the discussion of the technical challenges, the paper identifies promising research paths relying on interdisciplinary approaches combining classic dependability and system monitoring techniques with advanced machine learning that can predict failures but, importantly, also provide an explanation why a failure is imminent.