Architectural Run-time Models for Performance and Privacy Analysis in Dynamic Cloud Applications

Abstract

Building software systems by composing third-party cloud services promises many benefits such as flexibility and scalability. Yet at the same time, it leads to major challenges like limited control of third party infrastructures and runtime changes which mostly cannot be foreseen during development. While previous research focused on automated adaptation, increased complexity and heterogeneity of cloud services as well as their limited observability, makes evident that we need to allow operators (humans) to engage in the adaptation process. Models are useful for involving humans and conducting analysis, e.g. for performance and privacy. During operation the systems often drifts away from its design-time models. Run-time models are kept insync with the underlying system. However, typical run-time models are close to an implementation level of abstraction which impedes understandability for humans. In this vision paper, we present the iObserve approach to target aforementioned challenges while considering operationlevel adaptation and development-level evolution as two mutual interwoven processes. Central to this perception is an architectural run-time model that is usable for automatized adaptation and is simultaneously comprehensible for humans during evolution. The run-time model builds upon a technology-independent monitoring approach. A correspondence model maintains the semantic relationships between monitoring outcomes and architecture models. As an umbrella a megamodel integrates design-time models, code generation, monitoring, and run-time model update. Currently, iObserve covers the monitoring and analysis phases of the MAPE control loop. We come up with a roadmap to include planning and execution activities in iObserve.