Success Story: How an Open RAN PlugFest Cut Mobile Network Power Consumption by 10 Percent
December 02, 2022
Unless you’re a network engineer, you probably don’t think about how much energy the mobile network infrastructure that connects our smartphones and other gadgets consumes. But if you did, you’d be shocked. According to the International Energy Agency, global data transmission networks consumed 260-340 terawatt hours (TWh) of electricity in 2021. That’s roughly the same amount of energy that England, the country, consumed in the same year (333.2 TWh).
There lies one possible explanation as to why your mobile service bill has remained roughly the same over the last 10-15 years – because operators have to pass along those energy costs.
“Energy costs for mobile networks as we know account for up to 40% of the total cost,” says Eng Wei Koo, Director of Technology at Keysight Technologies. “The radio access network really accounts for more than three quarters of what we're looking at from that 40%.”
Energy is such a big expense because the RAN infrastructure currently supports peak traffic levels at all times in order to deliver the quality of service we expect. So, whether it’s the middle of the workday or the middle of the night, mobile network RANs are drawing the same amount of power.
Meanwhile, the average cost of a kilowatt hour in the United States has almost doubled since 1990, steadily rising from ¢6.57 to ¢11.18. With no end to those increases in sight, members of the O-RAN Alliance – an organization dedicated to making radio access networks (RANs) more open, intelligent, and interoperable – held a global PlugFest in the Spring with the goal of designing and testing disaggregated green network infrastructure that can adjust power consumption according to real-time traffic data.
In other words, when demand is high and the network needs to run at high capacity, requisite power will be made available. When demand is low, less power will be delivered.
“We used live mobile network traffic such as the peak traffic, which occurs during the commuting hours or after school hours, and then we used the quiet times when most of us are asleep,” says Nikoleta Patroni, Open RAN Senior Engineer at Vodafone.
“Based on these patterns, we managed to utilize the different power management tools and techniques to make sure that whenever the traffic is low, we can change the power state setup of our solution,” she continues. “And whenever there is high peak traffic, we would change the power state again to make sure that we can support this traffic regarding the standardized approach that we used.”
The 2022 O-RAN Alliance PlugFest occurred at the University of Utah’s POWDER laboratory, a facility dedicated to mobile networking experiments. It was attended by engineers from the Vodafone Group, Intel, Keysight Technologies, Radisys, and Wind River Systems.
The PlugFest was the first phase in a three-phase project to create a common approach to evaluating, testing, measuring, and monitoring the energy consumption of disaggregated Open RANs. From there, steps can be taken to adjust the amount of power delivered to parts of the RAN infrastructure based on need.
“In order to adjust the power dynamically based on the capacity need, you need to know how much power you're consuming, and you need to do it in an automated way,” says Gil Hellmann, vice president of telecom solutions at Wind River.
“You cannot connect the power analyzer to every cell site,” he continues. “The first phase is to understand, first of all, can we measure, successfully and accurately, the power consumption in a programmatic way, meaning through software?”
To accomplish this at scale without the limitations of traditional hardware-based measurement tools, the PlugFest team created what amounts to a digital twin of a 5G Open RAN based on traffic trend data from real-world Vodafone networks. They ran this traffic through a virtualized small cell 5G base station (gNB) running in an O-RAN Cloud (O-Cloud) built on and orchestrated through the Wind River Studio Cloud Platform (Figure 1). The gNB consisted of a containerized O-RAN Distributed Unit (O-DU) edge server and O-RAN Central Unit (O-CU). The entire stack executed on Intel® Xeon “Ice Lake” processors.
Figure1: The Wind River Studio Cloud Platform is designed to simplify the management of distributed cloud networks while ensuring customers maintain the high-quality user experience needed. (Image credit: https://www.windriver.com/resource/studio-operator-capabilities-cloud-platform-product-overview)
The digital twin was simulated using tools from Keysight’s open radio access network architect (KORA) portfolio (Figure 2), including the O-RAN Energy Savings test solution. Comprised of an emulated O-RAN Radio Unit (O-RU) with user equipment emulation capabilities, a core emulator, and the PathWave Test Automation Framework, the Energy Savings test solution provided stateful emulation of the digital twin based on the Intel processor’s P- and C-states, the former of which scales down frequency and voltage to conserve power while the latter either reduces or completely turns off select functions.
Figure 2: The KORA portfolio of solutions is designed to assist in testing an O-RAN's interoperability, security, conformance, and performance across its whole lifecycle.
Simultaneously, Keysight’s PA2203A IntegraVision power analyzer (Figure 3) was used to capture power-related measurements in real-time.
Figure 3: Keysight's PA2203A IntegraVision power analyzer, capable of providing power parameter measurements and real-time displays of current, voltage, and power traces across 4 channels. (Image credit: https://www.keysight.com/us/en/product/PA2203A/integravision-power-analyzer-4-channels-3-phase-ac.html)
“We can emulate the subscribers, the traffic profiles, and the network to look at mobility scenarios and the kind of services that yourself and myself basically perform when we connect our mobile devices,” Koo says. “One of the key challenges when we start to look at optimizing the network for energy efficiency and savings is, for example, there are multiple power saving states in the Intel processor. As you start to look at being more aggressive on getting the processors into some highly efficient energy saving states, you basically switch off certain capabilities and functions within the CPU cores, including switching off some CPU cores when they're not being used.
“But as you start to get more aggressive with the power management scheme, it takes a longer time for the processors to switch on certain capabilities and switch on certain CPUs to recover to the level where they can start serving the peak traffic that we are emulating,” he continues.
“When there are certain power management techniques being implemented, the Keysight emulation solution will need to be able to adapt accordingly and say, ‘Well when you start to look at saving power, when you start to look at maybe reducing the frequency of certain CPU costs, how would the subscriber sessions and the services that we're emulating together with other subscribers behave and perform?’”
According to Hellmann, the measurements acquired “through software are pretty much accurate and the same as the measurements from physical equipment.” That said, the PlugFest team’s repeated testing and fine-tuning of the real-life traffic scenarios provided by Vodafone yielded a reduction in the power consumption of high and low mobile traffic periods by 9 percent and 12 percent, respectively.
Fine-Tuning Open RAN Power for the Future
Now that the PlugFest has concluded, work to measure and quantify the effects of RAN power consumption adjustment more deeply is underway. For example, if the network is adjusted from full power and speed to a low-speed, deep power saving mode, how long would the transition take and would quality of service suffer?
Different spectrums are also being tested during this second phase to determine the viability of RAN power adjustment in use cases beyond indoor small cells. Given the long-term scope of this work is much broader, Hellmann says wider network capacities need to be examined to “get a full characterization of how things behave.”
Looking to the future, the third and final phase of this green RAN initiative will require the definition and implementation of standards. This is a critical component of the initiative given the open, multi-vendor nature of the Open RAN ecosystem, and the fact that operators like Vodafone plan to have 30 percent of their mobile sites running on Open RAN infrastructure by 2030. Hellmann believes that standardization work could complete in early 2024.
And as energy costs increase and regulations emerge in response to the climate crisis, power-saving standards for an industry with the energy footprint of a small country couldn’t come at a better time.