Aviz Service Node outlines BlueField-3 DPU offload for line-rate visibility

Aviz Service Node running on NVIDIA BlueField-3 offloads control-plane handling and line-rate user-plane steering from CPUs to the DPU, aiming to keep subscriber-correlated visibility while improving tool efficiency and scaling traffic optimization.

Research Overview

The blog describes performance and scaling pressures in telco and large-scale networks as data volumes and AI-era workloads grow. It argues that CPU-based packet processing for traffic management and observability can add infrastructure overhead and reduce efficiency as traffic increases.

To address this, the post positions DPU acceleration as a way to process critical functions closer to the network fabric, with subscriber-aware visibility and traffic control designed to scale without continuously adding server CPU capacity.

Key Findings

According to the blog, BlueField-3 provides programmable acceleration for both control-plane and data-plane tasks performed by Aviz Service Node. It also describes subscriber-coherent steering to route traffic to the appropriate probes, rather than sending all traffic to all tools.

The post states that moving these functions off the CPU layer reduces CPU dependency, enables line-rate data-plane processing, and balances probe capacity through subscriber-aware routing. It also describes the approach as extensible for future capabilities such as DPI, AI-based flow policies, and service chaining.

Technical Breakdown

The blog states that Aviz Service Node uses BlueField-3 to process control-plane traffic, accelerate data-plane flows, and steer subscriber traffic toward analytics tools. It says control-plane packets including GTP-C, PFCP, and N11 can be handled on BlueField-3 ARM cores.

For user-plane traffic, the blog says GTP-U flows run through BlueField-3’s P4 programmable datapath at line rate. It describes the resulting “visibility fabric” as correlating subscriber control traffic with user traffic before sending the right traffic to the right probes to improve visibility fidelity.

Operational Impact

The blog connects DPU placement with operational outcomes for observability at carrier-grade scale. It describes CPU-to-DPU offload as enabling line-rate performance at the interface and reducing the gap associated with server-stack latency and processing overhead.

It also attributes lower infrastructure cost to reduced need for additional CPU cores as traffic grows, claiming operators can handle more traffic without scaling server CPU infrastructure at the same rate. The post further frames the approach as addressing multi-tenant AI-era needs for isolation, subscriber-aware visibility, and consistent performance across tenants.

Overall, the blog describes an Aviz Service Node architecture on NVIDIA BlueField-3 that moves traffic management and observability functions from server CPUs to a programmable DPU, with subscriber-coherent steering and line-rate user-plane processing. This Blog Signals brief is a fact-based summary of the vendor blog.