Towards runtime optimization of software quality based on feedback control theory

Abstract

The increasingly complex environments in which software systems are running today have made runtime software quality unstable and hardly in an optimal state, especially for those systems in open and dynamic environments, e.g. Internetware. In this paper, we explore the effectiveness of software cybernetics and feedback control theory in runtime software quality optimization. We propose a method of runtime quality optimization by using feedback control theory. Specially, we consider the problem of runtime optimization for a specific quality attribute, namely throughput, for Web-based systems. We design a double-layer feedback control model for the problem and implement the runtime optimization control method. In the method, runtime feedbacks are collected and used by the control model to adjust related control parameters. The experimental study has demonstrated the effectiveness of software cybernetics and feedback control theory in runtime quality optimization.

Self-Tuning of Software Systems Through Goal-based Feedback Loop Control

“…PID controller synthesizes the benefits from P, PI and PD controllers and overcomes the drawbacks such as P controller's low stability, PI controller's high overshoot, and PD controller's slow response [11]. In the domain of software tuning, PID controllers have already been applied successfully in scheduling in real-time operating systems [13], automatic stress and load testing tools [14] and performance optimization [7].…”

“…Earlier we have reported a framework to optimize software quality at runtime [7]. The self-tuning method proposed there was based on low-level control parameters.…”

Self-tuning of software systems through dynamic quality tradeoff and value-based feedback control loop

“…Earlier we have presented a framework to optimise software quality at runtime (Chen et al, 2009). The self-tuning method proposed there was based on low-level control parameters.…”

“…PID controller synthesizes the benefits from P, PI and PD controllers and overcomes the drawbacks such as P controller's low stability, PI controller's high overshoot, and PD controller's slow response (Franklin et al, 2006). In the domain of software tuning, PID controllers have already been applied successfully in scheduling in real-time operating systems (Steere et al, 1999), automatic stress and load testing tools (Bayan and Cangussu, 2008) and performance optimisation (Chen et al, 2009).…”