Ericsson and Verizon back look-aside for Cloud RAN

Joint position paper says lookaside better for energy efficiency and system flexibility. Verizon has just deployed its first Ericsson vRAN site.

Ericsson has stated in a paper that is jointly branded with Verizon that it recommends lookaside acceleration for Cloud RAN L1 workloads.

The conclusion to a position paper says, “Ericsson and Verizon have concluded that, based on the current technology landscape, common Cloud infrastructure with Selected Function Hardware Acceleration currently offers the best path forward to deliver on energy efficiency, ecosystem support and flexibility at reduced complexity.” Selected Function Hardware Acceleration is Ericsson’s preferred term of lookaside acceleration.

The paper preceded a Verizon announcement that it has deployed its first vRAN site from Ericsson, as part of a drive to get to 20,000 vRAN sites deployed. However it already has over 10,000 vRAN sites deployed with its other RAN partner Samsung.

How to process low layer radio may seem an esoteric point, but it is not irrelevant to operators looking to Cloud RAN – the question impacts on cost, but also on how open an ecosystem can be. Ericsson said the choice of acceleration architecture impacts energy efficiency, programmability to enable multi-vendor ecosystems on common infrastructure, integration complexity, and TCO.

SoC companies offer complete in-line PHY acceleration programmed onto a PCIe card which offloads the whole L1. Lookaside acceleration sees the L1 still largely running on a CPU but only the LDPC functions are passed to a PCIe card via an API called BBDEV. This is the solution that is used for AMD (Xilinx) and Intel’s FPGA cards.

Opinions vary as to which method is preferable.

 

Want to know the Truth about Open RAN? Sign up for our upcoming webinar. Stats and expert analysis on the true state of Open RAN.

 

Vodafone, for instance, announced in October that it would be working with Qualcomm and Marvell, as well as EdgeQ, on products with Samsung and Nokia. These chip companies are committed to L1 inline acceleration, stating it is a more efficient use of CPUs. Rakuten is working with Intel on moving to its next generation of chip which, because of its new manufacturing method, is able to move from a lookaside accelerator card to providing on-CPU acceleration of L1 FEC.

But Ericsson has chosen to continue to back the lookaside design as its preferred option.

Look-aside and inline acceleration. Click for full size.

The paper said, “The ultimate benefit of Cloud RAN is the flexibility and scalability it provides to MNOs. The wide array of deployment options allows operators to choose hardware and infrastructure that best suit their needs, budget, and business model. Choosing the right architecture and configuration of hardware and acceleration can help MNOs reap the full benefits of Cloud RAN.”

As such, the position paper comes down on the side of lookaside acceleration – Ericsson prefers the term Selected Function Hardware Acceleration – on the grounds of energy efficiency and flexibility.

On energy it says, “A Full L1 Accelerator card can save CPU core consumption, however, it requires a separate PCIe Card to be inserted in the
server, which creates considerably more power consumption than a standard network interface card (NIC).
“With Selected Function Hardware Accelerator, there is an opportunity for application design to use a larger pool of available CPU cores efficiently; with tighter integration of selected accelerators within the CPU, the potential for further efficiency improvements.”

On system flexibility it says, “A Full L1 Accelerator card requiring [sic] software specifically developed for that specific hardware component which will
make disaggregation more challenging. Such cards often require software specifically developed for their hardware or chipset, which eliminates the possibility to create a common cloud compute infrastructure across the network and increases the risk of fragmentation.
“Additionally, in cases where L1 software is provided by the accelerator card supplier, the industry would need to take on added integration complexity to maximize the benefit of their features and to be able to diagnose operational issues as they arise.
“As such, using Full L1 Accelerator cards poses a challenge in terms of upholding some of the key principles in optimal Cloud RAN scenarios – namely, the strong desire for portability and flexibility.”