The power grid twitches every minute, and a software timer is behind it
For years, the frequency of the power grid has dipped at the start of every minute. Researchers in Jülich found the cause: thousands of backup batteries running a self-test at the same second.
What happened
Every minute, at exactly second zero, Europe’s power grid twitches. Researchers in Jülich have solved the mystery: the cause lies in the software of backup batteries.
The grid operator TenneT had seen it in Germany for years. At the start of every minute the grid frequency, the 50 hertz at which the current alternates, drops by a few thousandths of a hertz. It was not dangerous, but it was strikingly regular. A team led by the physicist Dirk Witthaut of Forschungszentrum Jülich went looking, reasoning that such a tight pattern only appears when many devices run on a clock and draw power at once. ‘It really was years of detective work,’ says Witthaut.
The breakthrough came on the research centre’s own campus. The team placed meters at some 80 points and found the pattern only in buildings with a battery-backed uninterruptible power supply, or UPS, which keeps computers running when the power fails. In a 1.5-megawatt UPS, the battery drew a short spike of current at the start of every minute, visible as far as the 110-kilovolt grid connection. The manufacturer confirmed that its battery management runs a self-test every minute. And because these systems set their clocks over the internet, thousands of devices test at once.
The study appeared in Nature Energy on 30 October. According to Dr. Web, the pattern shows up in data from twelve grids on five continents, from Iceland to South Korea and Australia. In Ireland, where data centres with large batteries use about a fifth of all electricity, the dip grew fourfold between 2016 and 2021, from about 3 to 12 millihertz. Single dips there now go beyond 20 millihertz.
Reports differ on which batteries are involved. Jülich’s press release speaks of large-scale battery storage checking its state of charge every minute; the study itself and trade press such as ingenieur.de describe self-tests in UPSs, such as those in data centres. After the researchers got in touch, the manufacturer released a software update that removed the fixed minute in one battery block.
It quickly became clear to us that the cause must be something controlled by the clock.
Dirk Witthaut of Forschungszentrum Jülich, in its press release of 30 October 2026
The background
How the grid stays in balance
A power grid has to generate as much electricity as it uses, at every moment. The frequency gives away how that is going: when devices ask for more than is coming in, it sinks below 50 hertz and reserve plants step in automatically. The heavy turbines of gas and coal plants act as a shock absorber, because their weight keeps them from slowing down easily. That damping is called inertia. Wind and solar farms connect through inverters and do not spin along, so inertia is falling.
According to the researchers, that is the main reason the dip is getting deeper: the same spike of demand leaves a bigger dent in a grid with less inertia.
The grid already had a rhythm
A fixed rhythm in the frequency is nothing new. Electricity is traded by the hour, and power plants ramp up or down in steps on the hour while demand changes gradually. As a result, grid operators have seen deviations around every full hour for years. On 10 January 2019 the frequency fell to 49.8 hertz at 20:02 UTC, partly for that reason: the largest deviation since 2006.
The minute dip is much smaller, but it comes sixty times an hour. In the Nordic countries, where hydropower keeps the grid in balance, the reserves respond to every small deviation. The turbines there slow down and speed up every minute, and wear out faster. The grid operators have already tightened their rules.
How thousands of clocks fall into step
Computers set their clocks over the internet, using the Network Time Protocol (NTP). That is exactly what you want, because a log entry or a payment is then accurate to a fraction of a second. But it also means that ‘every minute at second zero’ is the same moment everywhere. There is no standard for when a battery tests itself; Witthaut suspects most manufacturers independently decided that 60 seconds made sense.
Software engineers know this as the ‘thundering herd’: many programs doing the same thing at the same moment. The well-known fix is called jitter, a small random delay. In 2015 Marc Brooker of Amazon Web Services showed that retries without randomness keep arriving in waves, and with some randomness spread out.
Our analysis
Every manufacturer made a sensible choice, and together they made a spike
Nobody did anything wrong here. A battery that checks every minute whether it still works is sensible, and so is a clock kept in step over the internet. The problem only appears in the sum: thousands of devices that each made the same choice behave like one big machine drawing power at the same second every minute. The coverage tells it as a grid story, but it is a software story: a timer in a device in a basement shows up in the high-voltage grid.
The nuance is in the size. According to Dr. Web, a dip of 3 millihertz in the European grid stands for some 80 megawatts drawn at once, far from the limits at which grid operators raise the alarm. The risk lies in the growth: the grid’s inertia is falling, while the number of devices with an internet clock is rising. Our expectation: heat pumps, chargers and inverters will follow the batteries, and sooner or later grid operators will require such devices to spread their scheduled tasks.
The lesson applies to any organisation that runs software on many devices or schedules tasks at fixed times. For every scheduled task, ask: what happens if everyone does this at the same moment? A random delay of a few seconds costs nothing and takes out the spike, whether the load is electricity or a server.
What stands out
- Solved with meters on the home campus A puzzle that had kept grid operators guessing for years was solved with some 80 meters in the centre’s own buildings.
- The same second everywhere From Iceland to Australia the dip starts right after the full minute, because the devices set their clocks over the internet.
- Ireland shows where it is heading Where data centres use a fifth of all electricity, the dip grew fourfold in five years.
- Even an oven joined in A smart oven with an internet clock showed the same one-minute pattern whenever it was run on a timer.
What it means for you
Software on many devices behaves like one big device. A scheduled task at a round time can turn into a spike.
- Ask which tasks run at round times Backups at midnight, reports at nine, checks on the full minute. Have your colleagues or your supplier draw up that list.
- Spread them with a little randomness A random delay of a few seconds or minutes takes out the spike, and the task still does exactly the same thing.
- Measure the peaks, not the average A server or integration that is quiet on average can grind to a halt on the full minute. Look second by second around those times.
Where we come in · Development
We build software that doesn’t jump on the full minute
In custom software that runs on many devices or works on a schedule, such as IoT for machines, sensors and meters or integrations between your systems, we spread scheduled tasks and retries with a little randomness. That way not every device asks for power, data or attention in the same second.
Which tasks run on the hour in your organisation, and what if they all do it at once?
Tell us which of your software runs on a schedule. We’ll think it through with you and always reply within one working day.