Decentralized Resource Discovery, Prediction, and Composition for Inter-Domain Cloud-Edge Orchestration

The rapid growth of intelligent services across theCloud-Edge Continuum (CEC) is transforming the way computationand data are managed. As services are increasingly deployedacross multiple independently operated domains, new orchestrationchallenges are emerging, especially in environments whereprivacy, policy differences, and infrastructure diversity makeconventional centralized solutions impractical. These limitationsare particularly evident in Inter-DIMO (Distributed IntelligentManagement Orchestrator) scenarios, where orchestration musthappen across domains that operate with limited mutual visibilityand trust. To support intelligent coordination in such settings,several key challenges must be addressed: real-time resourcediscovery without centralized control, anticipating how resourceswill behave under changing conditions, and dynamically composingservices while respecting domain-specific constraints. Resourceprofiling and advertising represent crucial first steps,enabling resources to be described and shared in ways thatsupport discovery, prediction, and orchestration. The challengebecomes even greater at the edge, where unpredictability, limitedvisibility, and restricted resource availability are common. Toaddress these issues, current research is increasingly turning tomachine learning. Graph-based learning is being explored toenhance semantic discovery and matching, time-series modelshelp forecast availability under highly dynamic conditions, andreinforcement learning techniques support adaptive, distributedservice composition. Building on these approaches, orchestrationmust integrate resource profiling and advertising to ensure resourcesare not only found and predicted, but also communicatedin a way that enables scalable and decentralized coordinationacross 6G domains. Lightweight metadata exchange, decentralizedcoordination protocols, and abstract semantic models arebeing prioritized to promote scalability and interoperability indiverse, fast-changing environments. The forthcoming researchwill investigate how privacy-preserving learning, semantic-awareservice matching, and runtime adaptability can improve orchestrationframeworks. The broader goal is to advance towardautonomous, scalable, and resilient orchestration techniques thatcan meet