Aviz and NVIDIA Implement Subscriber-Aware Load Balancing on BlueField-3 DPU
Aviz Networks and NVIDIA have collaborated to implement Subscriber-Aware Load Balancing (SALB) on the BlueField-3 Data Processing Unit (DPU). This integration allows mobile operators to effectively manage control- and user-plane traffic per subscriber, enhancing visibility and reducing Central Processing Unit (CPU) usage while maintaining scalability.
Background
Mobile operators frequently encounter challenges in managing and correlating subscriber traffic effectively across various performance and monitoring systems. Control-plane and user-plane packets can often become disassociated, complicating performance analysis.
The Subscriber-Aware Load Balancing solution from Aviz, which operates on the NVIDIA BlueField-3 DPU, addresses this issue by correlating traffic per subscriber at line rate. This capability allows operators to maintain comprehensive visibility and optimize probe usage across large-scale environments.
Importance of Subscriber-Aware Load Balancing
SALB plays a crucial role by ensuring that control- and user-plane traffic remains intertwined for each subscriber. This ensures accurate analysis of performance data and minimizes issues related to traffic separation, which can create inefficiencies.
Mobile operators benefit from the capabilities of SALB as it maintains subscriber context and balances traffic distribution across monitoring tools, improving accuracy and decreasing overall system overhead.
Development of Subscriber-Aware Load Balancing
Work by Aviz and NVIDIA resulted in the SALB functionality embedded within the Aviz Service Node functioning on the BlueField-3 DPU.
- Control-plane packets: Managed by the DPU's ARM cores for precise processing.
- User-plane packets: Context is extracted and managed through P4 tables in the data path for line-rate efficiency.
This design approach enables coherent subscriber traffic steering while achieving high throughput and reduced CPU demand.
BlueField-3 DPU Architecture’s Role
The BlueField-3 DPU operates in a mode configured for the Aviz Service Node. Control and user packets are processed differently, enhancing overall efficiency and throughput.
The architecture supports proactive programming through the use of NVIDIA’s P4 Software Development Kit (SDK) to efficiently manage packet processing by pre-installing flow rules.
Performance Validation
Aviz conducted performance evaluations comparing setups using the BlueField-3 DPU and traditional x86 servers.
| Feature / Metric | BlueField-3 DPU (Setup 1) | x86 Baseline (Setup 2) |
|---|---|---|
| Hardware | BlueField-3 DPU (ARM + P4) | Dual-Socket Xeon (2.3 GHz) |
| Network Interface | Dual-port 100 GbE (200 Gbps) | Intel E810, 2×100 GbE |
| Traffic Generator (TG) | Keysight (Ixia) | Keysight (Ixia) |
| Control Plane | Processed by ARM cores | Processed by x86 cores |
| Data Plane | Processed by P4 Fastpath | Processed by x86 cores |
| Throughput (100 Gbps) | 1 ARM core + P4 datapath | ~20 x86 CPU cores |
The results indicated that one DPU port attained peak performance of 100 Gbps, while the x86 counterpart required nearly 20 CPU cores to achieve similar throughput, showcasing gains in processing efficiency.
Effects on Total Cost of Ownership and Power Efficiency
| Category | 100 Gbps Scenario | 400 Gbps Scenario |
|---|---|---|
| Core Efficiency | BlueField-3 (1×100G port) replaces ~20 x86 cores | BlueField-3 (2×200G ports) replaces ~80 x86 cores |
| Power Consumption |
|
|
The findings underscore how offloading tasks to the BlueField-3 DPU not only minimizes Total Cost of Ownership (TCO) but also enhances performance efficiency per watt relative to CPU-only systems.
Joint Validation Results
Aviz and NVIDIA concluded that the Aviz Service Node's SALB function performs effectively on a single BlueField-3 DPU by utilizing its high-performing ARM cores alongside the P4 programmable datapath. This configuration supports scalable Deep Network Observability while ensuring consistent performance.
Statements from Industry Leaders
“Predictable, line-rate telemetry steering is foundational for AI-era networks. With NVIDIA BlueField-3 DPU, operators can offload subscriber-aware functions into a programmable data path, keeping control and user traffic coherent per subscriber while preserving CPU headroom for other workloads.” — NVIDIA Networking spokesperson
“With Subscriber-Aware Load Balancing on BlueField-3, we correlate control- and user-plane flows per subscriber and steer them at line rate. That lets operators balance probe capacity, preserve fidelity, and scale observability without scaling cores.” — Chid Perumal, CTO, Aviz Networks
Advantages for Operators
- Subscriber-coherent analytics: Ensures combined analysis for control and user planes for each subscriber.
- Improved tool efficiency: Prevents probe overload while ensuring full visibility.
- High performance per watt: DPU offload surpasses CPU-based options in throughput and efficiency.
Conclusion
The Aviz Service Node utilizing Subscriber-Aware Load Balancing on the NVIDIA BlueField-3 DPU enhances efficiency, accuracy, and scalability in network observability. This unified approach to control and user-plane traffic benefits operational costs and aligns with the requirements of modern mobile and AI-driven networks.