SALB & NVIDIA BlueField-3 Load Balancing Brief | Aviz Networks
Subscriber-Aware Load Balancing (SALB) on NVIDIA BlueField-3 DPU
Subscriber-aware traffic steering for full visibility at scale
November 3, 2025
Why This Matters
Mobile operators must correlate control- and user-plane traffic (GTP-C, PFCP, N11 vs GTP-U) to maintain full subscriber visibility. Traditional x86-based probe architectures struggle to keep traffic coherent per subscriber while scaling throughput efficiently.
Aviz solves this with Subscriber-Aware Load Balancing (SALB) — part of the Aviz Service Node (ASN) running on the NVIDIA BlueField-3 DPU. SALB ensures subscriber context is preserved and both control- and user-plane packets are directed to the right analytics tools — at line-rate performance.
What We Built Together
Aviz validated the ASN with SALB on the NVIDIA BlueField-3 DPU, achieving subscriber-coherent traffic steering at 200 Gbps line rate.
Control Plane
ASN processes GTP-C, PFCP, and N11 packets on the BlueField-3 ARM cores.
Hybrid Offload Model
Control handled in software; user flows handled in P4 hardware for maximum performance and minimal CPU load.
User Plane
Subscriber context extracted → programmed into P4 tables for DPU datapath processing.
Proactive Programming
ASN uses the DPU SDK to dynamically install fast-path rules.
Figure 1: Aviz Service Node Architecture on NVIDIA BlueField-3 DPU
Validated Results
| Metric | NVIDIA BlueField-3 DPU | x86 Baseline |
|---|---|---|
| Throughput | 200 Gbps (line rate) | ~6–7 Gbps for single x86 core |
| CPU/Core Efficiency | 1 ARM Core + P4 Datapath | ~40 cores for 200 Gbps |
| Power Efficiency | ~80% less power vs CPU-based processing | Requires 560W to process 400G of traffic. |
A single BlueField-3 DPU replaces 40 x86 cores while preserving fidelity — dramatically lowering TCO and power consumption.
DPU (ARM Core + P4 Datapath) vs x86 CPU Cores
Experience Subscriber-Aware Load Balancing on NVIDIA BlueField-3.
What Operators Gain
- ✓ Subscriber-coherent analytics – correlate both planes per subscriber.
- ✓ Balanced probe utilization – avoid overload and data loss.
- ✓ High performance per watt – offload CPU-intensive tasks to DPU.
- ✓ Design flexibility – run ASN inline or host-based.
Connect with our experts to explore how Aviz can simplify and scale your AI infrastructure.