Evolving from Traditional Systems to AIOps: Design, Implementation and Measurements
Shijun Shen, Jiuling Zhang, Daochao Huang, Jun Xiao
Abstract
AIOps (Artificial Intelligence for IT Operations) has been proved to be effective in quality improvement, cost reduction and efficiency improvement, and is considered as the ultimate solution for IT operation and maintenance. But for most enterprises, it is still challenging to evolve from traditional systems to AIOps. This paper reviews the development of IT operation and maintenance technologies in the past two decades, and introduces five abilities that a typical AIOps system requires, namely perception, detection, location, action and interaction. Focusing on these abilities, we propose a novel AIOps system called Proton. Proton adopts the layered design with interoperability services between modules, which makes it well compatible with traditional heterogeneous systems. We have implemented Proton with some key considerations including data categories, database cluster, service gateway and operation safety. Proton has been deployed in a large IT system environment with tens of thousands of devices, and the measurements reveal that the fault self-healing rate of Proton exceeds 80% for the scenario of server ping failure.