SK Telecom and Intel Demonstrate High-Performance Virtualized User Plane

With the arrival of 5G, network operators will be challenged on two fronts: Most obviously, their network will need to deliver performance that satisfies the potential of 5G's massive bandwidth, but they will also need to ensure that the network remains scalable and flexible, so they can innovate freely and respond to new commercial opportunities.

To prepare for this need for flexibility and scale, South Korean mobile operator SK Telecom has already virtualized its entire mobile core. That includes the Mobility Management Entity (MME), Serving Gateway (SGW), Packet Data Network Gateway (PGW). Much of the old and new equipment in the network has been migrated to a virtual network function (VNF), including the IP Multimedia Subsystem (IMS), Value-Added Systems (VAS), and management systems that are virtualized and integrated to the Management and Orchestration (MANO) system.

However, the company's implementation was based on Single Root Input/Output Virtualization (SR-IOV) and PCI passthrough (PCI-PT). This enabled it to meet its performance requirements, but came at the cost of hardware and software dependencies that mean the solution can only be used in appliance-like deployments. It wasn't able to deliver the cloud-like agility, scalability, and flexibility that NFV is capable of.

Virtualizing without compromise

To enable more cloud-ready agility, Intel and SK Telecom carried out a proof of concept together to investigate how the virtualized Evolved Packet Core (vEPC) can be implemented on standard hardware. The aim was to demonstrate the performance 5G requires, without introducing software and hardware dependencies.

The proof of concept solution is based on the Intel® Xeon® Platinum 8180 processor, which has 28 cores and 56 hyperthreads, enabling the performance and virtual machine density required. The secret ingredients were the hardware-assisted Network Interface Cards (NICs). To improve performance, the NICs use hashing to direct all the packets from a particular user device to the same worker core on the processor, first time.

Enabling High Performance

Tests were conducted using four different traffic profiles, with different levels of upload and download activity. For example, one user profile had 10% uplink traffic and 90% downlink traffic, which might represent someone streaming video or other media content. Another profile was more balanced, with 42.5 percent uplink traffic, and 57.5 percent downlink.

For packet rates of between 30 and 42 million packets per second, the tests consistently showed more than 200Gbps throughput with latency of 70 microseconds. This demonstrates that the solution can be used to meet the performance requirements of 5G without compromising on hardware independence.

To find out more, read the solution brief, Accelerating the Virtualized User Plane for 5G Readiness, which includes further details of the hardware tested, how the tests were conducted, and the results seen.