Intent-based Meta-Scheduling in Programmable Networks: A Research Agenda
Nanjangud C. Narendra, Ronak Kanthaliya, Venkatareddy Akumalla
TL;DR
The paper addresses the challenge of coordinating multiple schedulers in large, programmable networks to fulfill user intents in 5G/B5G and beyond. It proposes a two-level meta-scheduling framework grounded in intent-based management and active inference causal reasoning, separating global intent fulfillment at the CU from local scheduling at the DU and leveraging DAG-based models and Markov Blankets. A strawman architecture with a set of cognitive agents and a meta-scheduling coordinator demonstrates how intents can be decomposed, evaluated, and enacted across the O-RAN stack, with considerations for interoperability and TMForum-based reporting. It also outlines a research agenda covering modeling, elasticity, integration of active inference, O-RAN standard enhancements, and real-world deployment scenarios such as UAVs, V2X, and Industry 4.0. The work aims to enable sub-millisecond, conflict-free resource scheduling at scale, improving QoS and flexibility in future networks.
Abstract
The emergence and growth of 5G and beyond 5G (B5G) networks has brought about the rise of so-called ''programmable'' networks, i.e., networks whose operational requirements are so stringent that they can only be met in an automated manner, with minimal/no human involvement. Any requirements on such a network would need to be formally specified via intents, which can represent user requirements in a formal yet understandable manner. Meeting the user requirements via intents would necessitate the rapid implementation of resource allocation and scheduling in the network. Also, given the expected size and geographical distribution of programmable networks, multiple resource scheduling implementations would need to be implemented at the same time. This would necessitate the use of a meta-scheduler that can coordinate the various schedulers and dynamically ensure optimal resource scheduling across the network. To that end, in this position paper, we propose a research agenda for modeling, implementation, and inclusion of intent-based dynamic meta-scheduling in programmable networks. Our research agenda will be built on active inference, a type of causal inference. Active inference provides some level of autonomy to each scheduler while the meta-scheduler takes care of overall intent fulfillment. Our research agenda will comprise a strawman architecture for meta-scheduling and a set of research questions that need to be addressed to make intent-based dynamic meta-scheduling a reality.