On Autonomic Platform-as-a-Service: Characterisation and Conceptual Model

In this position paper, we envision a Platform-as-a-Service conceptual and architectural solution for large-scale and data intensive applications. Our architectural approach is based on autonomic principles, therefore, its ultimate goal is to reduce human intervention, the cost, and the perceived complexity by enabling the autonomic platform to manage such applications itself in accordance with high-level policies. Such policies allow the platform to (i) interpret the application specifications; (ii) to map the specifications onto the target computing infrastructure, so that the applications are executed and their Quality of Service (QoS), as specified in their SLA, enforced; and, most importantly, (iii) to adapt automatically such previously established mappings when unexpected behaviours violate the expected. Such adaptations may involve modifications in the arrangement of the computational infrastructure, i.e. by re-designing a different communication network topology that dictates how computational resources interact, or even the live-migration to a different computational infrastructure. The ultimate goal of these challenges is to (de)provision computational machines, storage and networking links and their required topologies in order to supply for the application the virtualised infrastructure that better meets the SLAs. Generic architectural blueprints and principles have been provided for designing and implementing an autonomic computing system. We revisit them in order to provide a customised and specific view for PaaS platforms and integrate emerging paradigms such as DevOps for automate deployments, Monitoring as a Service for accurate and large-scale monitoring, or well-known formalisms such as Petri Nets for building performance models.