Energy and Edison
Remember the incandescent light bulb, a universal metaphor for a brilliant idea, invented in 1879 by Thomas Edison. It was a great invention that provided artificial light and expanded opportunities including reading and learning pre-dawn and past dusk. Only one problem, it wasted 95% of the energy it consumed in generating heat, only 5% was used for light. This continued until the first decade of the 21st century, when we started getting serious about energy usage, its cost and impact on the one home we call planet Earth. Compact Fluorescent started replacing incandescent bulbs and shortly after LED bulbs became affordable and now provide light using a fraction of the energy of an incandescent bulb.
Similar to incandescent bulbs, mobile networks use a disproportionate amount of electricity to deliver what the customer actually wants. Largely because of the way networks have historically been provisioned for peak usage, 15% of the energy is used to transfer data and the remaining 85% is wasted as heat loss in power amplifiers, idling equipment when there is no data to transfer and other inefficiencies in the components of the network. This level of waste is unnecessary and the telecommunications industry needs to make it’s transition to more efficient methods of delivering service..
The energy consumption of mobile networks contributes to greenhouse gas emissions and climate change, not to mention the cost for the operator. An interesting fact is that Alexander Graham Bell, the inventor of the telephone, actually coined the term “greenhouse effect” to explain the world becoming warmer due to unchecked burning of fossil fuels.
Improving energy efficiency can help reduce the carbon footprint of mobile networks and mitigate their environmental impact. Plus, energy-efficient mobile networks are less likely to experience power outages which can disrupt communication services and impact the user experience. Many countries have regulations in place that require mobile networks to meet certain energy efficiency standards. Failure to comply with these regulations can result in fines and other penalties, not to mention damage to reputation.
Seize the Now, Accelerate Change
According to GSMA research, the telecommunications sector is responsible for 2-3% of global energy usage and its share is steadily rising. A European Union report expects this usage to reach 10% of global energy usage by 2034. As connectivity continues to expand from mobile phones, tablets & home entertainment devices to medical devices, cars, refrigerators and various other sensors/IOT devices, more throughput results in more operator equipment and energy. Even though 5G is better optimized than 4G (i.e. it pushes more bits/watt), more 5G sites are needed than 4G for the same coverage area as 5G operates at higher frequencies and provides more bandwidth. Most of the energy we use today is not a renewable resource and geo-political conflicts add to the steady increase in energy prices. A large European operator was recently forced to increase mobile usage prices, attributing it to the cost of its energy bill.
Various governments around the world have set energy targets to reduce greenhouse gas emissions. With the European Green Deal, Europe aims to be the first carbon-neutral continent by 2050. This will involve legislation limiting carbon emissions for all industries, including telecom. Re-thinking and re-designing networks for energy efficiency takes a change in mindset and time. Mobile operators will be well-advised to start making changes in the way networks are designed and operated, making energy usage an important metric when purchasing hardware/software systems.
As Matt Ridley states in “How Innovation works,” innovation is “turning ideas into practical, reliable and affordable reality.” In other words, taking core inventions and applying them to solve real world problems in a reliable and affordable manner.
Communication Service Providers have been cautious until now, in adopting virtualization and commodity compute/network architecture for their services due to the stringent resiliency and latency requirements of their service offering. Now that these technologies have proven to be equal or better in every measurable metric than traditional mobile network solutions, CSPs can take advantage of the elasticity and on-going gains (for e,g. Moore’s Law) offered by the commodity chips and hardware.
Intel introduced Speed Select Technology (SST) in the 2nd Generation Intel Xeon Scalable processors, which were released in April 2019. This technology allows users to configure the performance of individual processor cores, allowing for greater flexibility in optimizing workloads for specific use cases. C-states determine how deep a sleep an individual core can go into, whereas P-states determine the frequency of the individual core at any instant. Intel allows the management of core and uncore frequencies on a per-core basis. Uncore frequency is the frequency of the non-core parts of the CPU, such as the cache and memory controller. As increased baseband functions are run as virtualized applications bundled as Kubernetes microservices, on commodity servers, the Intel tools enable the optimization of energy via control of the C-states and P-states.
A typical virtualized RAN (Radio Access Network) is disaggregated into Centralized Unit (CU) and Distributed Unit (DU). Each of these units consists of various Kubernetes pods running as microservices. Not every pod has the same latency and frequency/power requirements. Today, most deployments have all these microservices running on cores that run at full power all the time i.e., C-states at C0 and P-states at P0. Mobile networks' throughput requirements are highly variable throughout the day and typically follow a pattern for e.g., peak hours and non-peak hours. Also, some of the functions performed by the pods of the CU/DU never need the highest power profiles provisioned by default.
Wind River has worked with various partners such as Vodafone, Intel, Radisys, and Keysight to exercise the capabilities offered by SST in plug-fests and proof of concepts, dynamically calibrating the C-states and P-states of individual cores based on workload needs. Intel also demonstrated managing P/C states, using Intel FlexRAN, saved energy usage. Both exercises yielded energy savings of around 20%. Wind River Studio Cloud Platform now includes the creation of specific power profiles and their application to workloads, resulting in energy savings. Also, what is not measured cannot be managed. Studio Cloud Platform will generate energy metrics per pod to keep track of the energy usage at a granular level. This can be aggregated to provide energy usage at an application (for example, CU/DU) level. These time series energy metrics can be used by Wind River Studio Analytics or any third-party analytics to graphically plot energy usage over time.
Energy use in telecom will continue to increase as we move to 56/6G etc. and use higher frequencies for mobile services needing more equipment. This necessitates a different way of thinking about provisioning. In the past, telecom networks were built for Mother’s Day scenario (busiest calling day of the year),but were wasting energy idling the rest of the year. Now with the elasticity provided by commodity servers and the ability to control energy usage on a per-core level, CSPs can and should do better.
We, as partners in the telecom industry, need to help mobile operators with every tool at our disposal to make our networks more energy-efficient, to find our LED moment. Imagine what Alexander Graham Bell would think about a 5G network that also proactively works to reduce energy consumption and avoid the greenhouse effect!
About the author:
Anil Purohit is a Senior Product Line Manager at Wind River