Reconfigurable and Dynamic QoS Provisioning in Optical DCN Architecture for Cloud-Centric Real-time Applications

The escalating demand for cost-effective and traffic-optimizing applications has placed significant strain on current electrical packet-switched Data Center Networks (DCNs), which are hampered by bandwidth limitations. Optical circuit-switched DCNs, with their advanced capabilities of high data rates and enhanced bandwidth, offer a promising solution to these challenges but are impeded by prolonged reconfiguration times. To address these constraints, various passive optical device-based solutions have been proposed. Furthermore, in mixed traffic environments, there is a pressing need for dynamic Quality of Service (QoS) provisioning. This paper introduces a novel reconfigurable and dynamic QoS-provisioned DCN architecture designed to tackle these challenges. The proposed architecture employs a heuristic algorithm that enables real-time network reconfiguration, optimizing resource allocation and reducing latency while ensuring QoS. The effectiveness of the architecture is demonstrated through comprehensive simulations and hardware-based implementations. The architecture, referred to as Passive Optical Data Center Switch (PODS), integrates an Arrayed Waveguide Grating Router (AWGR) with a controller unit. The controller manages the dynamic placement of packets in buffers according to service class, with the added capability of reusing buffers from other service classes. This approach enhances QoS provisioning and enables path reconfiguration via a loopback method, thereby mitigating blocking and congestion. The scalability, high throughput, low latency, service reliability, and power efficiency of the PODS architecture are validated under real-time traffic conditions. Performance comparisons with existing passive optical DCN architectures, such as PODCA, reveal substantial improvements in delay, throughput, and blocking probability. The framework was simulated using Python on the Google Colab platform within a Windows environment, yielding a 46.3% reduction in latency and a 5% improvement in network load compared to the PODCA architecture. Additionally, a laboratory test bench model comprising 7 Top of Rack (ToR) switches and Raspberry Pi modules was constructed. This small-scale model featured 112 communication links utilizing 16 different wavelengths and achieved optimal performance with 14 buffers per ToR. Implementation of the proposed algorithm resulted in an 18.3% reduction in blocking probability across the entire architecture and eliminated blocking in high-priority real-time traffic under full network load conditions. The findings demonstrate that the reconfigurable DCN architecture presented in this study significantly enhances QoS delivery, effectively meeting the evolving demands of cloud services and data-intensive applications.