Adaptive Bandwidth Allocation for Optical Interconnects in Cloud Computing Infrastructure

Abstract

Cloud computing infrastructure is increasingly constrained not by computation alone but by the capacity, latency, and energy behavior of the interconnect fabric that binds servers, storage, accelerators, and tenants into a coherent system. Optical interconnects have become central to this fabric because they offer high bandwidth, low loss, and distance-insensitive transmission, yet their static provisioning often lags behind the bursty and heterogeneous demands of cloud workloads. This paper examines adaptive bandwidth allocation for optical interconnects as a system-level problem that spans physical-layer flexibility, network control, resource governance, and sustainability. It argues that adaptive allocation should be understood less as a narrow optimization of link utilization and more as a socio-technical capability that shapes fairness, robustness, deployment economics, and environmental accountability. The discussion analyzes architectural foundations, including wavelength switching, optical circuit switching, and hybrid electrical-optical topologies, before turning to control-plane requirements, telemetry, and learning-based management. It then considers structural trade-offs among latency, capacity, energy, and cost, and examines governance questions that arise when bandwidth becomes dynamically allocated across tenants, services, and regions. The paper also explores robustness, fairness, and sustainability implications, using cross-domain comparisons from wide-area traffic engineering and warehouse-scale computing. It concludes that adaptive optical bandwidth allocation must be designed with explicit policy choices, transparent operational metrics, and resilient fallback mechanisms if cloud infrastructure is to remain efficient, equitable, and environmentally credible.