Towards Fully Decentralized Self-Adaptive Reactive Systems

With the growing demand for real-time data originating from myriads of Internet-connected devices, the number of requests hitting today's computing infrastructures goes beyond what is manageable for operations and affordable for management. Coping with these challenges requires a modernization of the application architectures and the underlying infrastructures. The mobile nature inherent to modern communications and interactions requires a radical shift towards new computing paradigms that reflect the fully decentralized perspective of the emerging execution environment. To this end, the trend is to switch thinking from assembling components into systems to dynamically composing autonomous systems into systems-of-systems. Indeed, systems-of-systems possibly emerge dynamically as an opportunistic aggregation of systems available at a given time. Since these systems operate under highly dynamic conditions where both the entities and their interconnections are subject to continuous change, the traditional stability assumptions made on distributed systems' design are no longer valid. Indeed, the dynamic operating conditions introduce uncertainty, which may harm the dependability of the system. In order to guarantee the provision of dependable functionality in such an unknown, ever-changing execution environment, systems should be fluid and able to self-adapt their structure depending on the changing situation. This talk examines a set of principles and techniques facilitating the design and development of fully decentralized systems that leverage on self-adaptivity to mitigate run-time uncertainty. Specifically, the key objective is to efficiently and effectively provide engineers with proper abstractions to develop self-adaptive systems capable of being at the same time fluid, as well as dependable.