Title: Towards Runtime Adaptation of Actor Systems

Abstract: In this dissertation we focus on providing effective adaptations that can be localised and applied to specific concurrent actors, thereby only causing a temporary disruption to the parts of the system requiring mitigation, while leaving the rest of the system intact. We make the application of localised adaptations efficient through incremental synchronisation, whereby the specifier can strategically suspend specific parts of the system, whenever this is strictly required for ensuring that adaptations are effectively applied. We also study static analysis techniques to determine whether the specified incremental synchronisation is in some sense adequate for local adaptations to be carried out.
We thus identify a number of generic adaptations that can be applied to any actor system, regardless of its design and the code that it executes. We implement the identified adaptations as an extension of an existing Runtime Verification tool for actor-systems, thereby creating a RA framework for monitoring and mitigating actor systems. In parallel to our implementation we also develop a formal model of our RA framework that further serves to guide our implementation. This model also enables us to better understand the subtle errors that erroneously specified adaptation scripts may introduce. We thus develop a static type system for detecting and rejecting erroneous adaptation scripts prior to deployment, thereby providing the specifier with assistance for writing valid scripts. Although the static typesystem analyses scripts with respect to certain assumptions, we do not assume that the monitored system abides by these assumptions. We therefore augment our RA framework with dynamic checks for halting monitoring whenever the system deviates from our assumption. Based on this dynamically checked model of our RA framework, we prove type soundness for our static type system.