A service measurement framework for adaptive applications, adaptive infrastructures, and rapid problem diagnosis
Gregory B. Brewster, Joseph V. Elarde · 2006
Since the beginning of computer-based application development and implementation, the provision of stable application system performance has been and continues to be a significant and problematic issue. Poor performance is frequently a source of end user dissatisfaction and reflects on the entire quality of service delivered by IT. Maintaining stable application performance requires an effective ongoing management process that begins with accurate and objective measurements: including measurements of resource usage, application workload activity, and the service or response time delivered. However, end-to-end and component service measurements are not currently part of most Client/Server and Web application architectures. To address the issues described above, we develop a framework for end-to-end and component response time measurement, establish a flexible service measurement infrastructure/tool-set called ARTMS (Application Response Time Management System), and evaluate three uses of service measurement information. We make the following specific research contributions: (1) Design and implement end-to-end transaction and component correlation using a self-identifying tagged (embedded) packet based correlation communication. We consider multiple alternative designs, implement, and measure results. (2) Design, implement and test a class of adaptive algorithms that respond to service measurement feedback using potential extensions to the ARM API to facilitate communication between the application and the infrastructure. (3) Design, implement and test several new algorithms for rapid problem diagnosis using service and resource measurement feedback. In summary, this research examines and develops a standard methodology based upon the ARM (Application Response Measurement) API to implement ubiquitous end-to-end and component service measurements. Moreover, this research evaluates the methodology's effectiveness by exploring three exploitations of the measurements: Adaptive Applications, Adaptive Infrastructures, and Rapid Performance Problem Diagnosis.