Aviz Service Nodes add NVIDIA ConnectX support

Aviz Service Nodes now support NVIDIA ConnectX NICs for DPDK-accelerated packet processing, enabling wire-rate 100 GbE and higher throughput, hardware Virtual Extensible Local Area Network (LAN) (VXLAN) header stripping, and integration with Network Copilot; this affects visibility and deployment choices for enterprise networks.

Research overview

Aviz updated Service Nodes to add support for NVIDIA ConnectX network interface cards and to use Data Plane Development Kit (DPDK) for accelerated packet handling on x86 servers.

The change is presented for telco cores, data centers, and edge deployments that require line-rate packet inspection, subscriber correlation, and application-level metadata.

Key findings

ConnectX support enables DPDK-based offloads that deliver wire-rate processing at 100 GbE and above while reducing Central Processing Unit (CPU) usage compared with software-only processing.

Reported feature coverage includes GTP correlation, subscriber-aware metadata, application identification across thousands of signatures, packet deduplication, and ConnectX-specific Virtual Extensible LAN (VXLAN) header stripping.

Technical breakdown

Aviz integrates DPDK poll-mode drivers with ConnectX queues so Network Interface Controller (NIC) hardware handles packet engine tasks and processed outputs feed into Network Copilot for query and anomaly analysis.

Packet processing workflow

Service Nodes are deployed on x86 servers equipped with ConnectX NICs and integrate with existing packet brokers and infrastructure.

DPDK initializes poll-mode drivers to access high-speed queues and forward traffic for Deep Packet Inspection (DPI), GTP load balancing, and flow distribution at speeds exceeding 100 GbE.

VXLAN header stripping

ConnectX offloads removal of Ethernet, IP, User Datagram Protocol (UDP), and VXLAN headers to hardware engines, which reduces CPU usage compared with software stripping.

The vendor states hardware-based stripping can operate at line rates up to 200 Gbps with sub-microsecond latency and contrasts that with software stripping performance where a single core handles about 9 Gbps and roughly twenty cores are required to match hardware throughput.

Operational impact

The update supports deployment on commodity x86 servers from multiple vendors and, according to the blog, can lower Total Cost of Ownership (TCO) by more than 50 percent versus proprietary appliances.

Hardware-accelerated processing reduces CPU requirements for overlay handling and enables operators to scale throughput while maintaining deep traffic visibility and metadata correlation.

Enterprise IT and security leaders can assess ConnectX-enabled Aviz Service Nodes for wire-rate packet processing, hardware VXLAN stripping, and integration with Network Copilot for AI-based queries on commodity hardware. This “Blog Signals brief” is a fact-based summary of the vendor blog.