A multi-objective approach to model-driven performance bottlenecks mitigation

Software Performance Engineering (SPE) evaluates the key performance factors such as response time and utilization in the entire life cycle of software development. One of the important issues of software performance is bottlenecks that have not been investigated much till now in the process of SPE. Meanwhile, Bottleneck detection and mitigation in software modeling stage is quality-centered and cost eective. Layered bottleneck is a type of bottleneck that occurs in systems with layered services and aects its utilization more than at bottlenecks. The presented approach in this paper has selected Layered Queening Network (LQN) as an appropriate performance model to present and analyze the layered bottlenecks. The process of SPE from software model to performance model has been automatically implemented. Also, an optimization stage is added to nd the best specication of software model in a way that the strength of the bottleneck, the response time and the cost will be minimized. To assess the proposed solution, two recently proposed multi-objective gravitational search algorithms are employed. To evaluate the eectiveness of the applied algorithms, two well-known multi-objective algorithms: NSGAII and MOPSO are also applied to a case study, and a comprehensive comparison is presented.