Learning - based Adaptation Framework for Elastic Software Systems
Yingcheng Sun, Xiaoshu Cai, Kenneth A. Loparo · Proceedings/Proceedings of the ... International Conference on Software Engineering and Knowledge Engineering · 2019
Adaptation is a concern for elastic software systems.Conventional methods like Brownout try to deactivate optional computation per request after decoupling the software into different components, to lower the workload when peaks occur.However, resource-intensive components are not always easy to isolate, and some software systems are even not separable.In this paper, we propose a new paradigm that provides each core and mandatory component a corresponding alternative component, with similar function but lower resource consumption, and use reinforcement learning in a feedback loop control for the selfadaptation process.We modified the widely used benchmark RU-BiS to make it weakly coupled and add alternative components.Experiments show that our framework dramatically improves both the efficiency and effectiveness of self-adaptation.