How to Prevent Power Outages in Data Centers [Tech Talk]

The Lifeline of Data Centers, Backup Power Facilities
Power Supply Configuration with UPS and Backup Generators
Even Perfection Cannot Guarantee 100% Reliability

Among server terminology, there is a word called 'Uptime.' It is a percentage figure that indicates how long a server computer has been running without shutting down since it was last rebooted. For example, a data center with 100% uptime means that it has not been turned off for even a single second since the power was turned on.

100% uptime is like the lifeblood of a data center. If the data center stops operating for even 0.02 seconds, the entire system collapses. In other words, no matter what happens, the data center must run 24 hours a day without interruption. This is why power supply is the top priority for data centers.

Data Centers Consuming the Power of a Power Plant

Kakao Data Center. The photo is not related to any specific expression in the article. [Image source=Yonhap News]

However, a data center is a complex structure equipped with hundreds to tens of thousands of computing devices and wiring. It consumes 100 to 200 megawatts (MW) of power at once. Since this amount of power is almost equivalent to that of a large power plant, ensuring power supply to a data center 'no matter what happens' seems nearly impossible.

Therefore, all data centers install 'backup power facilities' within the building. In fact, they install not just one but two. These backup facilities are essentially the precious lifeline of the data center. The backup power systems of data centers are broadly divided into 'Uninterruptible Power Supply (UPS)' and 'backup generators.'

The Lifeline of Data Centers, Backup Power Facilities

An engineer maintaining the UPS battery system inside a data center. The UPS is an emergency measure that activates immediately in the event of a power outage to save the servers.
[Image source=YouTube capture]

Both UPS and backup generators are backup systems that operate when external power cannot be drawn. Typically, data centers consume more than 100MW of power at once, so ideally, power is brought from nearby power plants.

However, there is no guarantee that such systems operate at '100%' capacity 24/7, 365 days a year. Even a slight problem with a power plant, transmission facility, or transformer can cause power supply interruptions at any time. Whenever this happens, switches inside the data center switch over, immediately activating the backup power systems.

While both UPS and backup generators supply backup power, their detailed roles differ slightly. First, the UPS immediately supplies power to the server computers as soon as a power outage occurs. However, the UPS has limitations. Generally, the UPS distributes backup power stored in secondary batteries to the computers. It is more like an 'emergency protection device' to keep the suddenly powered-off computers alive as much as possible.

Meanwhile, while the UPS takes emergency measures for the servers, somewhere else connected to the data center, the 'backup generator' is running hard. These backup generators consist of diesel, gas generators, or electric motors. Being large equipment, they can generate power equivalent to a power plant, but they require a 'warm-up time' until they reach 100% operational capacity. The UPS is a system that buys time until the generator can fully function.

100% Perfection Is Impossible

Data centers are equipped with large gas turbines and diesel generators to prepare for power outages. The photo shows a data center generator from the American company Caterpillar. [Image source=Caterpillar]

Although data centers guarantee power supply through dual backup facilities, achieving 100% uptime completely is still very difficult. For example, in October 2022, South Korea's largest messenger system, 'KakaoTalk,' went down due to a data center fire. The data center facility rented by Kakao at that time also had UPS and backup generators. However, the fire occurred in the battery connected to the UPS, and eventually, the backup system itself shut down, making it impossible to prevent server stoppage.

Ultimately, there is no absolutely perfect system. IT companies that construct and manage data centers only prepare for emergencies through thorough dual or triple protection measures. They install heat sensors and fire prevention devices on all power delivery equipment, install special fire response systems on batteries vulnerable to fire, and even install UPS or backup generators redundantly.

Installing a single data center costs astronomical amounts. This cost is not only for high-performance CPUs or GPUs but also for the pure 'engineering difficulty' involved. Today, IT companies invest in numerous protection systems, including UPS, generators, fire-fighting facilities, and monitoring systems, to build safer data centers.

Because of this, data center construction costs soar to trillions of won or more, but compared to the potential loss of revenue and brand value caused by unexpected server downtime, it is still a manageable investment.