Sustainable energy management in data centers is one of the important topics of discussion today. According to American wise minds Jennifer Switzer and Barath Raghavan, information batteries can be an additional factor for responsible use of energy.
The world needs to get rid of the use of fossil fuels as soon as possible to stem the climate crisis. Yes right? Every company has a responsibility to implement clean and sustainable energy. Especially companies with large, energy-intensive data centers. For them, information batteries could offer a solution. Have absolutely no idea what information batteries are? We looked into it and came to this fascinating conclusion: they’re not really batteries.
Green, sustainable energy as standard
The need for renewable, clean and green energy is as great as it is necessary in light of the growing climate crisis. Solar and wind energy are logical solutions, but science faces countless challenges. One of them is the storage capacity of that energy. Another problem is that renewable energy sources are not always available. No one has yet succeeded in generating solar energy at night.
Energy-intensive structures, such as data centers, require a constant supply of electricity. During the day there are no problems and wind and sun can generate their share of the electricity to power an array of batteries. But it is not always enough, and especially at night there is a problem. Data centers therefore often still need fossil fuels to keep the equipment running.
Green data centers: why more horsepower doesn’t necessarily use more power
One solution that many researchers are working on is batteries with a larger storage capacity. For example, Google traded its St. Ghislain data center diesel generators for batteries. Yet capacity, along with skyrocketing costs, remains a major stumbling block to the eventual replacement of fossil fuels.
Information such as battery
The question put Jennifer Switzer, a doctoral student at the University of California San Diego, and Barath Raghavan, an assistant professor in the computer science department at the University of Southern California, into thinking mode. They came up with a theory of using data as a form of energy conservation. ‘Information batteries’, they called it, although that flag doesn’t quite cover charging. ‘Time-shifting’ comes closer. Shifting over time to doing computer calculations. The premise is noble: tackling the renewable energy availability problem.
With that question in mind, Switzer and Raghavan put their brains into storm mode. “If computers need energy to perform calculations, is the stored data a form of stored energy? Then information is like a battery,” the researchers said.
Time shift as a saving
“Within five years, as much renewable energy will be needlessly lost as the city of Los Angeles uses in a year,” Raghavan said in an interview with Science Daily. “We found that if we can predict what calculations a computer will need to perform, we can make those calculations happen in advance. And at a time when renewable energy is widely available.”
“The results of the calculations, the data or information, are stored and retrieved only when the computer needs those calculations. Because storing data uses much less energy, less fossil fuel is needed, which equates to a net saving from this time shift,” he continues Raghavan.
Google as an example
The idea of time displacement is not new. In 2020, Google already tested it by moving workloads in data centers. The goal was to perform translation tasks for Google Translate at times when renewable energy was widely available.
The researchers are excited to pick up another product from Google to prove the utility of their proof-of-concept. “YouTube encodes more than 700,000 hours of videos in different resolutions every day. Because these are predictable actions, they can be moved to moments when the renewable energy is almost inexhaustible. The encrypted files can then be easily stored on the server.”
A battery or not?
By shifting workloads and thus realizing energy savings, you are not yet creating a battery. Still, the researchers like to refer to their concept as information batteries. And somehow they have a point too. “A battery converts one type of energy into another. For example, using energy obtained by gravity or compressed air to move something else. Or to store data,” says Raghavan. He sees the stored calculations as a form of stored energy, which you use again at a later time.
Estimation is the key word in the paper by Raghavan and Switzer. Not all tasks qualify for this approach. Still, there are plenty of data centers that may have certain workloads or computations done in advance. A company like Netflix, like YouTube, could convert content for later subtitling. The same could be true when training machine learning (ML) algorithms. A scientist can train the machine to perform certain workouts at an energy-efficient time.
Microsoft will install Windows 11 updates if there is enough sunlight or wind
The planner at work
In order to make an estimate, the information battery works much like the planners in the operating system of a smartphone. These schedulers ensure that the data stream flows optimally through the processor and other chips.
In the case of the information battery, the manager has three components: a price indicator, a calculation engine (to determine which tasks should be performed in advance), and a scheduler.
To determine which tasks can be performed in advance, the scheduler weighs information from the price indicator and the calculation engine by an information battery. In Raghavan and Switzer’s model, the information battery manager weighed the indicators against each other every five minutes. When the electricity price fell below a certain limit, the manager had a series of calculations made and the results saved.
Save up to 30 percent
According to the researchers, the model was effective in reversing the need for ‘grid power’, as they called it. Even when the scheduler did not weight the indicators properly, the estimate often turned out to be correct and the tasks were scheduled at an energy-efficient time. In this way, companies could save up to 30 percent of energy costs.
In a typical data center, calculations can be estimated up to an hour and a half in advance with 90 percent accuracy.
“In a typical data center, calculations can be estimated up to an hour and a half in advance with 90 percent accuracy,” says Raghavan. “With a slightly more moderate forecast of around one hour, that data center could store 150 MWh. That’s significantly more than most grid-scale battery storage devices. Such batteries quickly cost $50 million.”
How much is it?
The authors do not state how much it would cost to operate an information battery. However, it will in any case be quite a bit smaller than the current infrastructure. Furthermore, it is implemented in software so that it can be optimized by the various parameters.
“The key to the correct implementation of the information battery is to realize that it is not a general solution. However, it is very effective for many common workloads in many enterprises and data centers,” the researchers conclude.
In the US, data centers use about two percent of all electricity, and that number will no doubt grow. In that respect, information batteries can become an affordable alternative to colossal and very expensive ‘real’ batteries. At least when renewable energy sources are used even more and more efficiently.