Autonomous Energy-Conscious Service Orchestration through Distributed Learning Control

Abstract

The fast growth of cloud-edge computing infrastructures and data-intensive intelligent applications has resulted in significant growth of system energy usage that has caused dire needs of scalable and sustainable orchestration of services. The current orchestration mechanisms are highly centralized and rule based which restricts their scalability, expertness and flexibility in heterogeneous and dynamically evolving environments. In an attempt to overcome these shortcomings, this paper presents a proposal of an autonomous energy-aware service orchestration framework, which relies on distributed learning control. The suggested solution is a decentralized orchestration intelligence in which local agents of learning are placed inside distributed executing nodes, which are able to make autonomous and cooperative decisions, but it does not rely on a centralized point to be controlled by a central controller. All agents dynamically control the placement of the services, scaling, and migration through an energy aware learning policy that also takes into account the resource utilization narratives, service latency, and quality of service (QoS) constraints. A collaborative reward model approach is presented to support the goal optimization of local interests with system-wide system energy efficient outcomes. The usefulness of the suggested framework will be tested by means of massive simulations of a large-scale cloud-edge system where heterogeneous nodes and time-varying workloads are considered. With experimental trials, the distributed learning-based orchestration can greatly decrease the total power usage, but keep QoS adherence and low service latency, and better compared to centralized energy-conscious schedulers and heuristic solutions. Moreover, scalability analysis proves that the suggested approach can ensure the same level of performance with the growth of the system size, which is why it can be employed in large-scale data engineering frameworks. The findings are that distributed learning control can offer a viable and scalable solution to intelligent and energy-efficient service orchestration of next-generation cloud-edge systems.