For many Virtualized Network Functions, a key requirement is to rapidly classify the nature of the packet flow. Is it a VoLTE call? Netflix? YouTube? A video conference? A torrent file sharing? This step is generally called packet classification and needs to be applied in real-time to potentially thousands of flows.
Once the nature of the each packet flow is known, then a mobile network operator can apply virtualized network functions (VNFs) in real-time to manage or improve the packet flow, such as policy enforcement, video optimization, bit rate adaptation, transcoding, CDR generation for real-time billing, or legal intercept. As network loads increase, an NFV architecture will dynamically turn up hundreds of VNFs. So another key requirement in an NFV deployment is to apply load balancing of the traffic loads across a farm of virtualized functions.
The Challenge: Both packet classification, and the load balancing across a pool of VNFs, takes processing power.
Radisys recently introduced the A4700 compute module and T-Series Compact rack mount servers, both of which are based on the latest Intel® Xeon® E5-2600 v3 series processors technology. Radisys has also invested in optimizing the use of Intel’s Data Plane Developers Kit (DPDK) on our latest Intel platform offerings, resulting in industry-leading wirespeed processing performance for data plane processing applications. Radisys Intel platforms are a great solution for processor-intensive packet processing applications, like packet classification and load balancing between VNFs. But even further improvements can be achieved beyond a collection of Radisys Intel compute resources. The improvement starts with the A2470 intelligent switch in a T-100 Series Platform combined with the population of A4700 compute modules.
The A2470 intelligent switch is a unique combination of COTS switching and network processing silicon controlled by Radisys FlowEngine™ software technology. Essentially, the A2470 is more than a state-of-the-art switch delivering up to 400 Gbps I/O performance per blade. The A2470 with FlowEngine can also be configured to deliver wirespeed packet classification and associated load balancing to distribute the VNF processing across a T-100 Platform system full of A4700 Intel compute blades.
How does the A2470 architecture make DPDK deployments better? By moving packet classification and load balancing from the A4700 CPU blades into the A2470 intelligent switch and freeing up capacity within the A4700 blades to apply more of the DPDK processing power for the VNFs running on the platform. This increased overall DPDK processing capacity results in less power, space, and cost for a given level of VNF processing requirements.
DPDK performance on the new Radisys A4700 compute modules and our latest T-Series Compact rack mount server is exceptional. And for smaller deployments, Radisys Intel platform offerings will deliver industry-leading economics for DPDK applications.
But when you need to scale your Intel compute resources in larger NFV deployments, the combination of the A4700 compute blades with the A2470 intelligent switch in a T-100 Platform will deliver even higher levels of DPDK throughput and performance and lower cost-per-Gbps processing requirements.
Click here to read our recently-published DPDK benchmarking study.