The Cost of Formal Verification in Adaptive CPS. An Example of a Virtualized Server Node

Cyber-physical systems (CPS) are large scale systems highly integrated with the physical environment. Given the changing nature of physical environments, CPS must be able to adapt on-line to new situations while preserving their correct operation. Correctness by construction relies on using formal tools, which suffer from a considerable computational overhead especially if executed on-line. As the current system model of a CPS may change to adapt to the environment, the new system model has to be verified at run-time prior to its execution to ensure that the system properties are preserved. CPS development has mainly concentrated on the design-time aspects, existing only few contributions that support on-line adaptation. We undertake a practical exercise to research on the pros and cons of formal tools to support dynamic changes at run-time. We formalize the semantics of the adaptation logic of an autonomic manager (OLIVE) that performs on-line verification for a specific application, i.e., a dynamic virtualized server system. We explore the realization of the autonomic manager using formal tools based on CLTLoc to express functional and non-functional properties of the managed system. The on-line verification manager services requests from mobile clients that might require a change in both the running software components and server services. To establish if the adaptation preserves the temporal constraints provided in the specification, i.e., to decide whether a new client request can be serviced in the modified system, the on-line verification manager employs CLTLoc satisfiability checking. In this scenario, we then provide empirical results showing the temporal costs of our approach.