The digital transformation boom is forcing more and more companies to migrate their business processes to cloud-based virtual environments. Accordingly, ever more data centers are being built both for the provision of cloud services and for the own needs of individual organizations. The number of accidents in data centers is also growing. The top three reasons behind this have included power supply problems for many years.
According to the Annual Outage Analysis 2020 by the Uptime Institute, data center outages in 2019 were most often caused by problems with IT services (31%), data networks (30%), and power supply (15%). Whereas in the first two cases, the functioning of the data center can be quickly restored, the malfunctions in the power supply systems can lead to unpredictable consequences.
On March 9, 2021, a giant fire engulfed the OVH data center in Strasbourg. Throughout the night, more than a hundred firefighters were battling the blaze with varying degrees of success. The incident rendered 3.6 million web servers unavailable, including government sites in France, Great Britain, Poland, and Côte d’Ivoire. This is comparable to damages caused by major cyber threats like ransomware.
In a short video, OVH founder Octave Klaba revealed that the fire was caused by an uninterruptible power source serviced the day before. Firefighters, having explored the scene with thermal imagers, confirmed this information.
So far, no one undertakes to assess the damage that OVH has suffered, but it is, of course, not about money only as the reputation losses are no less critical.
It is necessary to minimize the likelihood of serious failures in data centers caused by power supply systems. It is required to assess the current and future needs for backup power, carefully consider the selection and location of the UPS, and do other things. Let us take a look at what to pay special attention to.
It is not always possible for a local electricity supplier to provide the data center with the power of the required voltage and frequency. These properties can float within a fairly broad range. Fluctuations of parameters may cause a disaster. Therefore, a good modern UPS must be able to work with the input parameters that have a wide range of voltages and frequencies while providing reliable protection of its load.
The double conversion system ensures that, no matter how the input voltage floats, the output will exactly match the required parameters. This is achieved by converting the input AC voltage to DC and then converting it back to AC voltage of ideal output characteristics. These devices are often called online UPS since they permanently remain connected to the network, act as a stabilizer, and can instantly enable the autonomous power supply during a power outage.
The performance of an uninterruptible power supply unit is determined by the efficiency at each of the input-to-output voltage conversion stages (rectifier and inverter). As a result, only the final figure is taken into account, which characterizes the difference between the net power of the load and the total amount of electricity consumed by the UPS together with the load.
An efficiency of 97% is considered excellent for UPS data centers. This efficiency is usually achieved with a fairly high load of devices. Still, UPSs in data centers, as a rule, operate with a load of 50%, which leads to a significant decrease in energy efficiency (up to 85%). Therefore, it is important that the UPS uses technologies to improve efficiency when operating under a load much lower than the nominal.
For example, the VMMS technology in modular UPS units allows you to improve efficiency by increasing the load of modules required to support the load, and by disconnecting unloaded modules from the operation. Some models additionally support the economy mode that may bring efficiency up to 99% for cases where the double conversion mode gets automatically activated only when necessary. Considering the period of the UPS lifecycle, such savings are well worth the purchase of a device with this functionality even at a price higher than for a plain-type model.
Modular system and hot-swap function
Few of the clients of data center development projects would invest in extra computing power to cover the expected growth. Economic point of view implies gradually enhancing data center performance as the load increases. This also applies to backup power systems, the capabilities of which must keep up with the rest of the scaling processes.
A modular UPS would be the right choice as it allows you to increase the backup capacity in steps of 50 kVA at any time by simply adding the required number of units. Some modern models of UPS for data centers are designed so that manipulations with individual units can be carried out in a hot mode — without disconnecting the load and affecting the performance of critical IT systems.
To build up safe fault-tolerant uninterruptible power systems, redundancy of modules according to N+1, 2N, or combined schemes is used. For such a system to work correctly, its modules must operate correctly while having no direct wired communication with each other. A similar principle is implemented, for example, in the HotSync technology
Types of rechargeable batteries
VRLA (valve regulated lead–acid) has long been the most widespread type of rechargeable battery in data center backup systems. These are sealed, valve-regulated lead-acid batteries filled with liquid or gel electrolytes. These devices are supported by any modern UPS, although technologically, they are gradually becoming obsolete. Customers often choose VRLA due to its low cost and high prevalence. However, the issue of operating costs often remains behind the scenes. In the case of VRLA, the costs are quite high due to the short lifecycle of these batteries and the need for regular maintenance and testing.
The second type of battery increasingly used in data centers is the lithium-ion type. They are significantly more expensive than the lead type, but their properties fully compensate for that. Whereas the VRLA energy density is 15 – 50 W/kg, the lithium-ion type enables 70 – 260 W/kg. In other words, Li-ion batteries will be 4 – 5 times lighter than lead-acid batteries of the same capacity. Often in the process of designing a data center, you have to take into account the maximum load on floor slabs of a building, and such arithmetic can play a role.
In addition, lithium-ion batteries withstand 2 – 3 times more recharge cycles (1000 versus 200 – 400) and can last 10 – 15 years, which is three times longer than lead-acid do. Even if initially, to save money, the data center backup power system uses VRLA batteries, you can consider switching to a more advanced solution before the next battery replacement cycle. Transformation costs can be minimized by purchasing a UPS compatible with both lead and lithium-ion batteries.
Finally, I should mention supercapacitors, which are also increasingly used in data center UPSs. The autonomous load time they provide is usually less than a minute, but this may also be enough to switch the data center to a backup power channel or start diesel generators.
Supercapacitors can be in service for up to 20 years. They do not require any maintenance, are fireproof, and charge literally in 1 – 2 minutes. In addition, they remain fully operational at temperatures ranging from – 40 to + 65 °C, which expands the scope of their application to the tasks that other types of energy storage devices cannot cope with. It is sad but there are not so many UPSs that can work with supercapacitors.
So, let me list the traits of a quality UPS: it should use double conversion voltage in operation, have a capacity of 95% or more, and work with different types of batteries in case of future upgrades. In addition, modular design and the ability to replace and add individual modules without disconnecting the load are very important. Finally, when choosing, it is always worth considering the reputation of the manufacturer. After all, saving on nameplates can also be very expensive.