Accelerating Forwarding Plane Performance

Intel Select Solutions for NFVI defines the minimum NFVI hardware and software for specific application use cases and sets the baseline for performance.

For example, the SKY-8101D 1RU dual-socket server based on 2nd Gen Intel Xeon Scalable processors offers 200Gbps of PCIe bandwidth per socket, for up to 400Gbs per 1RU. This is ideal for applications that are inline (for example BNG) where 100Gbps upstream and 100Gbps downstream capacity per socket maximizes the number of serviceable user sessions.

While testing the platform with the Intel Select Solutions for NFVI Forwarding Platform scripts, Advantech recorded 319 Gbps of vBNG throughput on the SKY-8101D server integrating Intel® Ethernet 800 Series Network Adapters, while overall server power consumption was less than 550W. When testing this configuration with simple Layer 3 forwarding, a platform throughput of 396 Gbps (99% line rate) was achieved with 256-byte packets.

Tests were also performed which make use of Enhanced Dynamic Device Personalization (DDP) on the Intel Ethernet 800 Series which allows reconfiguration of the network adapter’s internal packet filtering and processing pipeline on-demand to meet specific use case needs. This is achieved by adding new configuration profiles to the Ethernet controllers at runtime without resetting or rebooting the server using the Data Plane Development Kit (DPDK).

This configuration profile provides the ability to classify packet types and protocols inline and distributes the packets to specific queues on the device host interface. These queues can be assigned to specific cores on the CPU. This enables application performance scale with the number of cores in the system improving overall performance and utilization. It also improves latency because the packets end up automatically on the right worker core. For example in a BNG application it is vital for performance that all the packet flows from the same CPE are handled by the same core to effectively enable technologies like billing & HQoS.

In addition, SKY-8101D server configuration integrating the Intel® Ethernet Network Adapter E810-CQDA2 also provides unique Application Device Queues (ADQ) capabilities. ADQ technology improves performance scalability and predictability by dedicating queues to key workloads, delivering predictable high performance with dramatically reduced jitter.

This platform addresses general use cases for NFVI, focusing resources on I/O to provide the widest data path possible into each NUMA node. Optional components can be added to meet the requirements of specific use cases. Intel Optane DC persistent memory can be added to the configurations to provide massive memory resources that enlarge the pool of warm data that can be held in close proximity to the processor. User plane nodes are available in two primary configurations, with configurability to fine-tune the stack for specific solution needs: Base Node and Plus Node.

In addition, the reference architecture specifies system configuration parameters for the Intel Select Solutions for NFVI Forwarding Platform Controller Node.

The Advantech SKY-8101D is a verified Intel Select Solutions for NFVI Forwarding Platform and is designed for high throughput and reliability across the CommSP Network Infrastructure. It is built around the architectural and deterministic performance features of 2nd Generation Intel® Xeon Gold processors.

The system is made up of the following key components and technologies:

━ Balanced non-uniform memory access (NUMA) connectivity: I/O is evenly distributed across both CPU sockets. Workloads running on both sockets can benefit from direct access to network, storage (Non-Volatile Memory Express or NVMe drives), and Intel® QuickAssist Technology (Intel® QAT) acceleration. 

━ High throughput network adapters: The SKY-8101D offers four NUMA balanced PCIe x16 Gen3 slots. For Plus node validation, four dual-port (QSFP28) 100Gb Intel Ethernet Network Adapters E810-CQDA2 were installed in the system.

━ Data Plane Development Kit (DPDK) to accelerate packet processing in the data path. It also facilitates building packet forwarders designed to operate on general-purpose, standards based servers. 

━ Dynamic Device Personalization (DDP) is a programmable packet-processing pipeline provided by the Intel Ethernet 700 and 800 Series that supports on-demand reconfiguration of network controllers at runtime, enabling workload-specific optimizations to increase throughput and decrease latency. 

━ Application Device Queues (ADQ) is a capability first implemented in the Intel Ethernet 800 Series that provides dedicated queues to key workloads, enabling application-specific data steering, signaling, and rate limiting using an optimized application thread-to-device data path. The ADQ innovation increases application response time predictability, reduces latency, and improves throughput.

With RES, developers can get ahead of the curve and begin to test different NFV infrastructures on platforms destined for deployment closer to the subscriber in the access network, mobile edge and in customer premises (uCPE) as well as the network core and telecom data center.