An Object-Oriented Data Model for the Automation of Computer Performance Management

Abstract

Performance management is a complex, multifaceted discipline incorporating many activities. These include, but are not limited to, performance problem detection and diagnosis, performance tuning, capacity planning and workload forecasting. These activities have much in common; in particular, they share the same performance data requirements, and employ similar performance data interpretation techniques. However, performance management applications often replicate knowledge about performance data, its interrelationships and interpretation. This results in redundant storage of information, duplication of programming effort and loss of information. The resulting applications suffer a variety of difficulties, including drawing different conclusions from the same raw data; as well as potential synchronization problems, in that each application updates its own local model of system state at a different time. Recording performance data in a central repository may help with this problem. In this paper we explore the application of object-oriented data modelling to express the structure and complex interrelationships existing between performance measurement variables. We motivate object-oriented data modelling as a useful approach, and propose a small set of basic classes and methods to model computer performance data. Encapsulated in the data model are methods for the collection and management of real-time performance data, and methods that support a simple application for the diagnosis of real-time performance problems. We feel that the object-oriented approach provides a powerful and effective technique for capturing both the structure of performance measurement data, and common diagnostic techniques for the interpretation of performance measurement data.