Abstract

Traditionally, non-functional requirements (NFRs) are specified as measurable entities to permit evaluation satisfaction; however, NFR specifications quickly become obsolete because (1) NFRs are expressed in numbers, (2) architects specify them using the correct values at design time, and/or (3) providers are constantly improving their offer, in terms of functionality and quality of service (QoS). The computing-with-words approach has already been proposed to replace numerical NFR specifications, where natural language words denote fuzzy quality levels; unfortunately, current proposals provide only for design-time, stakeholder-defined translation of words as numerical ranges. We propose a mechanism to automatically and dynamically determine current numerical ranges of the fuzzy quality levels from the available data, without human intervention, whenever changes to component QoS specifications. Our main contribution is allowing architects to specify their requirements using words only once (at design time), and whenever providers change components QoS characteristics, automatically update those requirements to the new market view, enabling market-aware requirements. The approach was validated by measuring the number of times that necessarily a requirement had to be rewritten at runtime in order to get new operationalizations which replace the now older ones. We use a set of ten complex requirements, a dataset of 1500 actual Web services with precise measurements for nine QoS aspects, and a simulated offering variability. A Web-based prototype is also made available.