Syslog AI: An Artificial Neural Network as a Feedback Control for Network Infrastructure

Syslog messages were not designed to be analyzed by humans in a cognitive manner. These messages are highly technical and specific to the machine generating the operational data. Syslog messages are interpreted by trained humans specific to that domain. These messages are high-volume, fast-moving and oftentimes an afterthought, but can be the first touch when an incident occurs. Over time, not only can technologies and vendors change, but humans with that specialized knowledge become scarce. Here we established an artificial neural network and used machine learning to predict an outcome. We first identified and isolated a human workflow, changed that into a controlled feedback system to reduce variability and subjectivity, and then used the predictive engine to determine whether a syslog message needed to be acted upon, thereby alleviating the human touch and creating actionable intelligence. Managing the flow/control of syslog messages has always an uphill fight: what to exclude and making sure to include things. This session will help explain how we removed the human element from parsing the UNIX syslog messages and how we were able to determine true events that are actionable. We will also touch on how we use nontraditional protocols (GRPC) to “flip” how we monitor and troubleshoot our wireless service.