HPE and Qualcomm Technologies began analyzing the total cost of ownership for RAN, in depth, post Mobile World Congress 2019. They realized that, while RAN presented one of the most challenging targets for virtualization, it also promised the biggest return on investment if implemented for optimal economic efficiency.
What I remember most about MWC2019 was talking to many operators about the then relatively popular, shiny object called Open RAN or Virtual Radio Access Network (vRAN). It was like the unicorn at the event: noticeably desired by all, talked about by many โ but realized by none.
All the operators I spoke with wanted the dis-aggregation of hardware and software in RAN so that they could deploy virtualized RAN software on general-purpose hardware. This was the desired, much-needed end state, required to attain procurement flexibility, reduce capital expenditure (CAPEX) and operating expense (OPEX), avoid vendor lock-in, and facilitate innovation in 5G at the edge. Standards bodies and consortiums alike converged on the optimum hardware and software functions splits for various deployment scenarios. On paper, it all looked good: keynote speeches from top executives conveyed this message, meetings were held promising this message, and hands were shaken on the promise of delivering on the perfect end-state.
All was well. Or so it seemed to be at that time.
But the realist operators โ especially those in the business and operations side of the house โ knew that RAN was the most budget-intensive part of a telco network when it came to total cost of ownership. Given this, the desired end state presented operators with a huge mountain to climb. Furthermore, driven by the increasing spectrum bandwidth availability across bands, evolving massive MIMO coding and beamforming antenna techniques, associated baseband algorithm evolution, and the ever-growing traffic, the computational complexity of RAN baseband was only going to explode in 5G. All of this made operators seriously question the economics of RAN hardware and software dis-aggregation and deployment on general purpose compute platforms.
The total cost of ownership, defined as the CAPEX (amount spent buying hardware and software) as well as the OPEX (the amount spent operating the hardware and software), would probably prevent them from attaining this end state for a long time โ or perhaps never. Business justification to adopt open RAN was getting weaker by the minute.
Do they continue to be right? Not anymore.
Why? This is when Hewlett Packard Enterprise and Qualcomm began analyzing the topic in depth to determine that โ while RAN presents one of the most challenging targets for virtualization โ it also promises the biggest return on investment if implemented for optimal economic efficiency.
Achieving this goal of lowering TCO required re-thinking the foundations of how vDU platforms and the most compute intensive parts of the RAN software are architected and processed. Simply asking hardware vendors to sell their products at a loss in the name of decreasing the TCO was not a sustainable future. The industry needed to jump the tracks and think differently about vRAN and the Telco Edge.
The idea really, if you think about it, is quite simple. I would like to call it the brainchild of the collective thinking of my HPE team, with my Qualcomm colleague and co-author of this blog, Gerardo Giaretto, and his team.
First, take the very dense HPE ProLiantโข DL110 (1U form factor with 4 PCIe slots) and provide up to 5 (including the CPU) heterogenous programming environments, each quite powerful in their own capacity. Next, populate the PCIe slots with the ever-powerful RAN L1 processing environment with the Qualcomm X100 cards. Then, realize increased performance from the HPE and Qualcomm Inline acceleration solution โ while saving on power consumption and reducing TCO.
Making the magic happen
What were the steps to make this magic happen?
- To start with, inline processing provides vDU efficiency.
- Inline processing also provides heterogeneous programmable environments.
- These heterogeneous programmable environments allow the workloads to run on the silicon best suited for the underlying algorithms, providing unsurpassable system-level compute performance and efficiency. Here are some examples:
- High order 16-layer massive MIMO requires matrices manipulation with non-linear compute complexity
- Single Instruction Multi-Data (SIMD) native DSP is the most efficient for this type of workload.
- Next, the full off-load of the L1 provides the efficiency of statistical pooling gains coming from traffic multiplexing across more cells.
- In other words, the CPU has more bandwidth to process just the L2/L3 and pool activities related to multiple cells (i.e., multiple L1 environments).
- Elimination of single vendor technology lock-in. How?
- By compartmentalizing the implementation lock-ins in both silicon and software domains
- Providing interoperable programmable interfaces between these environments
- Decoupling the technological lifecycles of the main vRAN components
- Implementing the vision of a true Open RAN.
- Localizing most latency-critical processing.
- In other words, eliminating the low latency interactions over slower PCIe lanes (i.e., L1 processing between the CPU and a card in a PCIe slot in the look-aside architecture).
- Relaxing the latency and jitter targets for the CPU
- Enabling more liberal power and dormancy control policies in the host
- All leading to power savings for the operator.
- In other words, the next-generation 5G vDU is capable of finer control over performance and dormancy states in each of the programming environments of the architecture and localizes most latency-critical processing.
- In addition, it no longer needs to leverage the coarse and slow power control and dormancy mechanisms of PCIe. Such an approach further lowers power consumption in idle and low traffic scenarios, bringing the goal of zero traffic, zero power in vRAN closer to reality.
Even after we (HPE and Qualcomm) articulated this new architecture and the benefits of the next generation vDU, the vRAN software developer community was concerned that these programmable environments were closed. Be assured that these concerns have no basis. This is because the Qualcomm team has always taken these into consideration โ and are providing today โ a C-based software development kit (SDK) to allow software developers to re-program the X100 card to their heartsโ content.
So, there you have it โ open, programmable next-gen vDU with better performance and power. True realization of rainbows and unicorns!
Optimize your rollout of true Open RAN and vRAN
In summary, the next generation vDU from HPE and Qualcomm can achieve up to 60% TCO improvement for vRAN infrastructure. Because of this, operators will be able to realize their stated goals from MWC 2019 while eliminating vendor lock-in, in both silicon and software, and reducing their physical footprint by over 50% for high-density deployments โ not to mention the drastic reduction in overall power consumption.
Our hope is that the operators take advantage of these benefits to expedite their rollout of the true Open RAN and vRAN, getting closer to attaining the shiny object they heard about pre-pandemic and implementing quicker than they had previously thought.
And our final wish is that the 2023 Mobile World Congress is powered by this next generation vDU technology, and we see this technology become mainstream with the early adopters of 5G vRAN.
Join us at Mobile World Congress 2022!
Meet the authors:
Gerardo joined Qualcomm in 2007 and has since covered different roles in research, system design, standardization and product management. He recently was the product manager of Qualcomm Snapdragon X50, the first worldwide 5G Modem-RF Platform. Before joining Qualcomm Gerardo was part of the standardization team for Telecom Italia.
Gerardo received Bachelorโs and Masterโs degrees in Electrical Engineering from the University of a Padova and has been granted 55 patents.
The specific opportunities in her current portfolio include the deployment of the 5G technology, Open RAN, virtual RAN (vRAN) in particular, as well as MEC, Core Networks and Private 4/5G Networks. She leads customers from concept to delivery. Geethaโs leadership spans the management of the partner ecosystem for all vertical solutions as well as creation of inventive business models for customers, worldwide.
Prior to HPE, Geetha led large global R&D teams for Wireless Terminal Business Unit at Texas Instruments Inc. (TI) and wireless and switching telecommunication products at Nortel Networksโข for 23 years, managing hundreds of hardware and software engineers spanning over a dozen countries.
Geetha brings the experience of introducing innovative products to market, with inventive business models, which has resulted in proven system solutions that generated over $200M in revenues within the first 18 months and over $1B in 3 years.
Geetha holds a Masterโs Degree in Systems and Computer Engineering from Carleton University in Ottawa, Canada. She holds 4 US patents in Telecommunications Software, and several awards for leadership vision and people development.
You can reach out to Geetha on linkedin
Telco Industry Marketing
Hewlett Packard Enterprise
twitter.com/HPE_Telco
linkedin.com/company/hewlett-packard-enterprise
hpe.com/solutions
