An ontology-based framework for model-driven analysis of situations in data centers

Abstract

The capability to analyze systems and applications is commonly needed in data centers to address diverse problems such as root cause analysis of performance problems and failures, investigation of security attack propagation, and problem determination for predictive maintenance. Such analysis is typically facilitated by a hodgepodge of procedural code and scripts representing heuristics to be applied, and configuration databases representing state. As entities in the data center and relationships among them change, it is a challenge to keep the analysis tools up-to-date. We describe a framework that is based primarily on the principle of interpreting declarative representations of knowledge rather than capturing such knowledge in procedural code, and a variety of techniques for facilitating the continuous update of knowledge and state. A metamodel representing data center-specific domain knowledge forms the foundation for the framework. A model of the data center topological elements is an instantiation of the metamodel. Using the framework, we present a methodology for conducting a variety of analyses as a model-driven topology subgraph traversal, governed by knowledge embedded in the corresponding metamodel nodes. We apply the methodology to perform root cause analysis of performance problems in the domains of 3-tier Web and Info Sphere Streams applications. © 2013 IEEE.