Symbiotic and sensitivity-aware architecture for globally-optimal benefit in self-adaptive cloud

Abstract

Due to the uncertain and dynamic demand for Quality of Service (QoS) in cloud-based systems, engineering self-adaptivity in cloud architectures require novel approaches to support on-demand elasticity. The architecture should dynamically select an elastic strategy, which optimizes the global benefit for QoS and cost objectives for all cloud-based services. The architecture shall also provide mechanisms for reaching the strategy with minimal overhead. However, the challenge in the cloud is that the nature of objectives (e.g., throughput and the required cost) and QoS interference could cause overlapping sensitivity amongst intra- and inter-services objectives, which leads to objective-dependency (i.e., conflicted or harmonic) during optimization. In this paper, we propose a symbiotic and sensitivity-aware architecture for optimizing global-benefit with reduced overhead in the cloud. The architecture dynamically partitions QoS and cost objectives into sensitivity independent regions, where the local optimums are achieved. In addition, the architecture realizes the concept of symbiotic feedback loop, which is a bio-directional self-adaptive action that not only allows to dynamically monitor and adapt the managed services by scaling to their demand, but also to adaptively consolidate the managing system by re-partitioning the regions based on symptoms. We implement the architecture as a prototype extending on decentralized MAPE loop by introducing an Adaptor component. We then experimentally analyze and evaluate our architecture using hypothetical scenarios. The results reveal that our symbiotic and sensitivity-aware architecture is able to produce even better global benefit and smaller overhead in contrast to other non sensitivity-aware architectures.