INTEGRATING DPU-ACCELERATED SDN STACKS INTO CLOUD COMPUTE FABRICS

Abstract

The growing magnitude and complexity of cloud computing systems have posed serious challenges to conventional host-based networking systems. The data centre networks of today have limited scalability and performance due to the heavy computational load that general-purpose processors require to handle high packet processing rates, complicated virtualization architectures, and high-density east-west traffic patterns. Recent developments in programmable networking technologies have proposed data processing units (DPUs) that from host systems in addition to while enabling high-performance packet processing directly on programmable hardware. This review examines the integration of DPU-accelerated software-defined networking (SDN) stacks into cloud compute fabrics. The literature is reviewed to assess architectural frameworks, programmable data-plane technologies, and experimental performance evaluations related to accelerator-based network systems. Findings from earlier studies indicate that programmable networking devices substantially enhance throughput, reduce latency and lower host CPU usage in high-performance cloud settings. Packet processing pipelines implemented in hardware further allow better isolation and security of infrastructure in multi-tenant data centre systems. Nevertheless, there are still issues with programming abstractions, infrastructure observability as well as heterogeneous hardware platform interoperability. Future research directions include building advanced orchestration models, programmable telemetry and hardware-software co-design approaches that have the potential to support large-scale programmable networking infrastructures. The results imply that SDN structures accelerated by DPUs will play an increasingly significant role in the design of the next generation cloud computing architecture.