AI-driven Workload Management in Meta OS

Abstract

Properly leveraging resources within a continuum, and in particular the satisfaction of user requirements, requires deep understanding the workload itself and its interaction with the infrastructure in which it is executing. We present a novel approach to workload placement and execution within the cloud-to-edge continuum that leverages advanced AI capabilities. These capabilities allow the presented approach to optimize workload allocation, both in terms of resources used, and ability to satisfy explicit and implicit user defined service level requirements. We tested the approach against synthetic and real-world datasets, demonstrating the advantages over a baseline approach.