Requirements problem and solution concepts for adaptive systems
When we do Requirements Engineering (RE) for adaptive systems, are we doing RE
as usual? Do adaptive systems pose new challenges to RE? If yes, what kinds of
challenges are these? The talk gives preliminary answers to these questions.
The main argument of the talk, is that the requirements problem and solution
concepts for adaptive systems are different from the de facto standard problem
and solution concepts in RE, introduced by Pamela Zave and Michael Jackson in
their seminal 1997 ACM Transactions on Software Engineering and Methodology
paper, "Four dark corners of requirements engineering". In the talk, I argue
that the solution concept in adaptive systems RE is not a single specification
as in Zave and Jackson's requirements problem, but more specifications along
with evolution requirements that impose constraints on which specifications we
can switch between at runtime, whereby switching occurs when we observe that
the system has failed to satisfy requirements to the desired extent. I relate
this to key ideas in RE for adaptive systems, including monitoring, feedback
loops, probabilistic relaxation, fuzzy relaxation, and evolution requirements.