Modelling Uncertainty for Requirements: The Case of Surprises

The concept of Surprise has been used to model uncertainty for self-adaptive systems (SAS). The uncertainty in the environment of SAS demands the system to perform self-adaptation decisions, and therefore makes it hard to formulate, validate and manage their non-functional requirements (NFRs). A number of probabilistic measures exist that compute surprise to flag up situations of uncertainty for NFRs. A problem with these measures is that they don't give any information about how big or small the surprise is, and therefore lack support for quantification of the level of uncertainty and its impact on NFRs. The challenge here is to classify the size of surprise to better model uncertainty levels for NFRs. We argue based on classification, surprise can be used to identify failure situations for NFRs, and thereby support modelling of risks. In this paper, we propose a framework to allow for the classification of surprise. Based on this framework, we perform an analysis into risk for the NFRs. As a proof of concept, we have applied the framework to case of Remote Mirroring. The risk analysis is performed on both individual NFRs as well as combinations of them to demonstrate the framework's utility across a variety of tasks.