Ethernity Demo of 5G UPF Offload

As communications service providers continue to race toward initial 5G rollouts, they are recognizing that they cannot achieve the required benchmarks to enable true 5G performance without significant data plane acceleration. Offloading data processing and resource-consuming virtual functions to an FPGA-based SmartNIC is the most efficient means of achieving such acceleration and eliminates heavy loads on the CPU. It frees those resources for user sessions that require deterministic results for high throughput and low latency.

To that end, Ethernity offers its 5G UPF Acceleration solution, which offloads 5G User Plane Functionality software to Ethernity’s FPGA SmartNIC to accelerate 5G packet processing at the network edge, where high bandwidth and low latency are of key importance.

By offloading the data plane to Ethernity’s FPGA-based ACE-NIC100 SmartNIC, the solution provides extremely high bandwidth and low latency while saving CPU cores.

Service providers are seeking to complement and enhance their 5G networks by deploying Open UPF, in combination with edge connectivity, and are choosing to use FPGA SmartNICs for acceleration. With the ACE-NIC100’s small footprint and low power dissipation, the Ethernity solution is truly optimized for edge deployment, allowing service providers to boost performance, reduce networking overhead, and lower Total Cost of Ownership, as well as collocate the UPF with other services.

To demonstrate the performance and agility of such an offload solution, Ethernity recently created a demo video of CLI-implemented UPF software, offloaded to the ACE-NIC100. This particular version of the Ethernity solution is capable of connecting to any UPF-C that does not run over DPDK.  Included in the video are three different tests:

  1. I-UPF, in which a service table of subscriber information is created and the system is configured to handle hundreds of thousands of flows of data with classification and 4 million counters
  2. The conversion of 100,000 flows of data from IPoE with multiple tuples to GTP and back, QoS Marking, as well as per-flow MEF compliance policer and rate limiters
  3. sFlow, in which traffic type is monitored

The ACE-NIC100 performs these tasks well, showing high flow counts at high throughput with the required features to implement the 5G data plane efficiently and with high performance.



By Brian Klaff