Cost-Aware Resource Orchestration in Multi-Cloud Environments: A Framework for Intelligent Workload Distribution and Infrastructure Optimization

Multicloud deployments provide a lot of flexibility to enterprises, and while they can gain significant cost savings, doing so without compromising service levels is an open engineering challenge. In this paper, the authors propose a scheduling framework called Cost-Aware Resource Orchestration (CARO) to simultaneously minimize cost, maximize resource utilization, and enforce the service-level agreement (SLA) in a heterogeneous provider ecosystem. CARO is a combination of a Temporal Fusion Transformer price-forecasting module, a scheduling agent based on Proximal Policy Optimization, and an online SLA constraint monitor. Compared to static assignment to a single cloud provider, a 29.6% reduction in cloud spend per month was achieved over 12 months of field studies across AWS, Azure, GCP, IBM Cloud and Alibaba Cloud, and mean resource utilization increased from 59.4% to 82.1% without impacting SLA violations, which averaged below 3% through all load levels up to 120% of rated capacity. The one-way ANOVA revealed the following results: F(2, 177) = 312.7, p < 0.001, eta-squared = 0.78; effect sizes for all the pairwise contrast exceeded Cohen's d = 1.97. These results put CARO on the path to becoming an economical and reliable deployable solution for multi-cloud operations, backed by statistical validation.