Preemptive infrastructure maintenance through adaptive pattern search
Florin B. Manolache, Nathan Pitzer, Hanyu Yang, Tong Bian, Octavian Rusu · 2017
Since computing infrastructure has intrinsically a redundant logic, budget saving practices or understaffing of IT departments lead to an “emergency room” service paradigm. However, a treasure of information that could facilitate preemptive maintenance is buried in the logs and messages that are automatically generated by regular processes. Such information is typically ignored because of the shear volume of the logs in an enterprise environment containing hundreds or thousands of computers. This work presents a system which adaptively extracts the useful hints from large amounts of automatically generated data. This system has two components: a filter stack to extract raw information from logs and from automatically generated messages, based on generic criteria that can be easily configured; a tag based search engine that learns the expertise of a human operator, by propagating it into the configuration of the filter stack. The components of this system were developed and used at Carnegie Mellon University to transform some of the “emergency room” IT support into preemptive maintenance that increased server uptime and avoided data loss.