When we’re discussing vRAN and Open RAN topics, we’ll find ourselves facing some questions about the performance and cost. The highly common question is ” Can virtualized RAN deliver the same performance as traditional RAN? ” I see it’s a valid and logical question especially since no one will invest with at least a guaranteed TCO or Performance or Both.
What is the main challenge of vRAN and Open RAN?
Basically, vRAN and Open RAN are about RAN disaggregation which means splitting the RAN into different types of components RU, DU, and CU. At each component, there’ll be some RAN protocol to be hosted from the physical layer till PDCP and RRC. Then move some of these RAN components to the edge Datacenter of central DC to decrease the power consumption in the site, improve the mobility between different sites by centralizing the processing, and utilize the IT Cloud technologies which will reduce the TCO by reducing the Capex and Opex.
One of the most demanding workloads known to the cloud world is the 5G RAN, particularly the processing of Layer 1 network functions such as beamforming. And RAN Split has different options as per 3GPP, each option has its own pros and cons. The functional split adopted in Open RAN decides what proportion of these Layer 1 processes run on the radio unit or on the virtualized baseband (in an Open RAN architecture, the virtualized distributed unit or vDU). The O-RAN Alliance favors split 7.2, which places the majority of these L1 functions on the vDU which creates a very high processing burden for the servers and chips that underpin the vDU.
In current technology, a central processing unit (CPU) based on Intel x86 or on ARM cores cannot support L1 processing to the same performance as dedicated chips in a traditional RAN, without huge trade-offs in cost and power efficiency.
Most chip and systems vendors rely on accelerators – chips dedicated to a specific task, which offload that task from the CPU, boosting both performance and efficiency. All the major processor makers that are targeting their offerings at vRAN servers have piled into the market with accelerators like Intel, Marvell, Nvidia, and Qualcomm.
However the case is different when you’re speaking about the top RAN vendors like Ericsson and Nokia, Nokia and Ericsson initially based their base stations, whether virtualized or not, on their own processors and accelerators.
Conclusion
Whether, vRAN or Open RAN, It’s mandatory to understand how the layer 1 network functions will be offloaded, the integration between the accelerator cards which will offload some of all L1 functions and the main processing CPU, and the performance/cost of these integrations. Also, the type of acceleration that will be used in-line of lookaside which is provided by Intel, so far.
What is the difference between look-aside Acceleration and inline Acceleration ?
Moving the workload to the vDU to offer a good capacity for the fronthaul interface and make it open by following Split 7.2x is still the main challenge!