Knowledge Base K Models to Support Trade-Offs for Self-Adaptation using Markov Processes

Abstract

Runtime models support decision-making and reasoning for self-adaptation based on both design-time knowledge and information that may emerge at runtime. In this paper, we demonstrate a novel use of Partially Observable Markov Decision Processes (POMDPs) as runtime models to support the decision-making of a Self Adaptive System (SAS) in the context of the MAPE-K loop. The trade-off between the non-functional requirements (NFRs) has been embodied as a POMDP in the context of the MAPE-K loop. Using Bayesian learning, the levels of satisficement of the NFRs are inferred and updated during execution in the form of runtime models in the Knowledge Base. We evaluate our work with a case study of the networking application domain.