Aviz Networks details subscriber-aware load balancing on NVIDIA BlueField-3 DPU

Aviz Networks and NVIDIA have introduced Subscriber-Aware Load Balancing (SALB) implemented on the NVIDIA BlueField-3 Data Processing Unit (DPU) to assist mobile operators in correlating control- and user-plane traffic per subscriber. This approach aims to enhance the accuracy of subscriber traffic analysis while optimizing tool utilization and reducing Central Processing Unit (CPU) load in 5G network observability.

Research overview

Mobile Network Operators (MNOs) encounter complexities in maintaining subscriber-level coherence between control-plane packets (such as GTP-C, PFCP, and N11) and user-plane packets (like GTP-U), which typically traverse the core network separately. Without proper correlation, this separation risks diminishing context and analytic precision, while also increasing CPU demands and the potential for probe overloading.

To address these challenges, Aviz Networks developed SALB deployed on the BlueField-3 DPU, offering line-rate performance for comprehensive traffic steering and real-time correlation at scale.

Technical breakdown

SALB functions within the Aviz Service Node by associating control- and user-plane packets for individual subscribers, ensuring these flows are processed jointly. Control-plane traffic is handled by the BlueField-3 DPU's Arm cores, while user-plane packets are directed through the DPU's programmable P4 pipeline to sustain line-rate throughput.

The architecture configures the BlueField-3 in DPU mode, leveraging its dual-port 100 GbE interfaces and NVIDIA’s P4 Data Path Library Software Development Kit (SDK) to proactively install processing rules. This setup supports subscriber-coherent steering with high efficiency and low CPU utilization.

Key findings

Performance tests demonstrate that the BlueField-3 DPU achieves 100 Gbps throughput on a single Arm core combined with the P4 datapath, whereas a comparable x86 server setup requires approximately 20 CPU cores. This highlights the efficiency advantage in throughput and power consumption for the DPU-based solution.

Power consumption comparisons show the BlueField-3 operates at around 110 watts, saving approximately 50 watts compared to dual-socket Xeon CPUs at 160 watts for 100 Gbps scenarios, and an estimated 450 watts saved for 400 Gbps scenarios versus 560 watts typical of x86 systems.

Operational impact

Implementing SALB on the BlueField-3 DPU enables operators to maintain subscriber-aware analytics by ensuring all packets for a subscriber are consistently directed to the same analytical tools. This avoids probe overloading and preserves data fidelity.

Operators achieve improved tool efficiency and enhanced performance per watt, reducing computational overhead and infrastructure costs without sacrificing observability at scale.

Leadership perspective

Chid Perumal, CTO of Aviz Networks, stated SALB's capability to correlate control- and user-plane flows per subscriber at line rate supports balanced probe capacity and observability scalability without the need to increase CPU core counts.

This endorsement underscores the practical application of SALB in operational environments focused on optimized network observability.

Aviz Networks and NVIDIA jointly validated SALB’s operation on the BlueField-3 platform, confirming its scalability and suitability for deep network observability demands at line rate with minimal CPU overhead.

The combined technologies provide a scalable approach to integrate control- and user-plane correlation necessary for modern 5G network management.

This Blog Signals brief summarizes the vendor’s detailed presentation of Subscriber-Aware Load Balancing as deployed on NVIDIA’s BlueField-3 DPU, offering enterprise network decision-makers a factual overview of this method’s operational and efficiency benefits.