Understanding Data Centers from Logs: Leveraging External Knowledge for Distant Supervision

Data centers are a crucial component of modern IT ecosystems. Their size and complexity present challenges in terms of maintaining and understanding knowledge about them. In this work we propose a novel methodology to create a semantic representation of a data center, leveraging graph-based data, external semantic knowledge, as well as continuous input and refinement captured with a human-in-the-loop interaction. Additionally, we specifically demonstrate the advantage of leveraging external knowledge to bootstrap the process. The main motivation behind the work is to support the task of migrating data centers, logically and/or physically, where the subject matter expert needs to identify the function of each node - a server, a virtual machine, a printer, etc - in the data center, which is not necessarily directly available in the data and to be able to plan a safe switch-off and relocation of a cluster of nodes. We test our method against two real-world datasets and show that we are able to correctly identify the function of each node in a data center with high performance.