When people picture a datacenter's electrical resilience, the first image that comes to mind is usually the diesel generators roaring outside. But between the instant the utility supply drops and the moment those generators start, synchronise and stabilise their voltage, several seconds go by. In that gap, the only barrier between the failing mains and a total service outage is the UPS (Uninterruptible Power Supply), known in Spanish as SAI (Sistema de Alimentación Ininterrumpida).
The UPS is, literally, a bridge. It stores energy — usually in batteries — and delivers it instantly when the mains fails, keeping the load powered with no perceptible interruption until the generators take over. But its role is not limited to blackouts: every day it protects equipment against micro-outages, surges, voltage sags and electrical noise that, while never quite switching anything off, degrade power supplies and cause faults that are hard to diagnose.
In this article we explain the types of UPS that exist, the battery technologies that power them, how a system is sized for a technical room, which redundancy schemes are used and — the point that causes the most outages in practice — how it is maintained and tested so it responds on the day it is really needed.
The UPS as a Critical Bridge
The best-known function of a UPS is to cover the generator start gap. A modern genset needs between 8 and 15 seconds to start, reach running speed and take on the load with stable voltage and frequency. Without an intermediate battery, those seconds would be enough to shut down servers, storage arrays and network gear. The UPS absorbs that interval by delivering energy from its batteries with a transfer time that, in datacenter-grade equipment, is zero milliseconds.
Its second function, less visible but equally important, is power conditioning. The mains is not a perfect source: it suffers millisecond micro-outages, voltage spikes and dips, transients caused by nearby large loads switching, and high-frequency noise. These disturbances never actually cause a shutdown, but they punish equipment power supplies, shorten their lifespan and can corrupt writes in progress. A good UPS filters all of that and delivers a clean, regulated waveform to the load.
That is why the UPS is not an accessory but a central link in the power chain: mains → UPS → PDU → equipment, with generators as long-duration backup behind them. If the UPS fails, neither the generators nor the mains matter, because the load loses its immediate source of energy.
UPS Types: Offline, Line-Interactive and Online
Not every UPS offers the same level of protection. The IEC 62040-3 standard classifies units into three main topologies according to how they deliver energy to the load:
- check_circle Offline or standby: the load is fed directly from the mains and the UPS only kicks in when it detects a fault, switching to battery in a few milliseconds. It is the cheapest option, but that switch involves a micro-outage and it does not condition power continuously. It is designed for non-critical equipment: workstations, small routers or consumer electronics.
- check_circle Line-interactive: it adds an automatic voltage regulator (AVR) that corrects highs and lows without drawing on the battery, extending its life. It still switches to battery during a real outage, but handles everyday voltage variations far better. It is the typical choice for small server rooms and SME environments.
- check_circle Online double conversion: this is the datacenter standard. Mains energy is first converted from AC to DC and back to AC, so the load is always fed from the inverter, never directly from the mains. The result is zero transfer time during an outage and full isolation from any upstream mains disturbance.
The practical difference is significant: in an online UPS the load never "sees" the mains directly, so a micro-outage, a spike or a frequency distortion never reaches the equipment. That permanent isolation is what justifies its higher cost and its slight efficiency penalty versus simpler topologies. Reference manufacturers such as Eaton, Vertiv or Schneider Electric build their technical-room ranges on this topology.
UPS Types Comparison Table
The following table summarises the protection level, relative cost and typical use of each topology:
| Type | Protection | Cost | Typical use |
|---|---|---|---|
| Offline / standby | Basic (switches with micro-outage) | Low | Non-critical equipment, workstations |
| Line-interactive | Medium (AVR voltage regulation) | Medium | SMEs, small server rooms |
| Online double conversion | Total (zero transfer, isolation) | High | Datacenter, critical loads |
| Rotary / flywheel | High (brief kinetic backup) | High | Bridge to generator, no batteries |
Battery Technologies: VRLA, Lithium-Ion and Flywheel UPS
A UPS stores its energy somewhere, and that somewhere is usually a battery bank. The chosen technology drives cost, space, weight and — above all — long-term reliability. There are two major chemical families and one mechanical alternative:
VRLA lead-acid batteries (Valve Regulated Lead Acid) have been the industry standard for decades. They are cheap, extremely well proven and need no electrolyte maintenance. Their weak point is lifespan: between 3 and 5 years under ideal conditions, which shrink dramatically with temperature. They also take up a lot of space and weigh considerably, no small factor when we are talking about banks capable of sustaining hundreds of kW.
Lithium-ion batteries have become the preferred option in new installations. They cost more upfront, but offer decisive advantages: a lifespan of around 10 years, half the space and weight for the same energy, much greater tolerance to ambient temperature and management systems (BMS) that monitor every cell. Over the datacenter's lifecycle, their total cost of ownership is usually lower than lead-acid despite the higher initial outlay.
There is also a chemistry-free alternative: the rotary or flywheel UPS. Instead of batteries, it stores kinetic energy in a mass spinning at high speed; when the mains fails, that inertia is converted back into electricity for a few seconds. It offers only a handful of seconds of autonomy — not enough for a long blackout — but exactly what is needed to cover the generator start, with no batteries to degrade or recycle and a very high tolerance to temperature. It is often seen paired with generators or as a complement to battery banks.
Sizing: kVA, kW, Autonomy and Efficiency
Sizing a UPS correctly is an exercise that combines power, time and growth margin. The first step is to calculate the total load it must support. Here it helps to distinguish two magnitudes that are often confused: kVA (apparent power) and kW (real power). The relationship between them is the power factor: modern UPS units work at unity power factor (kVA ≈ kW), but older gear required 0.8 factors that reduced the usable power available.
The second parameter is autonomy: the minutes the UPS must sustain the load on its batteries. In a datacenter with generators, there is no point sizing for hours; it is enough to comfortably cover the genset start plus a safety margin, typically 5 to 15 minutes. That margin allows a failed start to be absorbed and retried, or an orderly shutdown to begin if the generator were not to start.
The third factor is the growth margin. A UPS installed at 90% of its capacity leaves no room to expand the room, and running so close to the limit also reduces its efficiency and life. The usual approach is to leave headroom that allows growth without replacing the unit and, in large installations, to opt for modular architectures where power modules are added as the load increases.
Rule of thumb:
Size by the real measured load, not by the sum of the equipment nameplates, which are almost always overrated. Add the expected growth margin and a reserve for redundancy, and check that the UPS keeps good efficiency across the usual load range, not just at full power.
Finally, efficiency has a direct impact on the electricity bill and on the datacenter PUE. Double conversion introduces unavoidable losses, but today's UPS units exceed 96-97% efficiency in online mode. Many also feature an ECO mode that partially bypasses the double conversion to push efficiency above 99%, at the cost of a slightly longer transfer time; it is used with judgement, only where mains quality allows.
Redundancy: N+1 and 2N
A single UPS, however good, is a single point of failure. That is why datacenters deploy several units in redundant configurations. The two most common schemes are:
- check_circle N+1: one more unit is installed than is strictly needed to sustain the load. If "N" is three UPS units to cover demand, four are installed; the failure or maintenance of any one of them does not affect service. It is a good balance between availability and cost.
- check_circle 2N: the whole system is fully duplicated into two completely independent power paths, each able to take on 100% of the load on its own. Dual-corded equipment connects to both paths. It is the maximum-availability scheme, typical of Tier III and Tier IV installations.
The choice between the two depends on the committed service level and the budget. We analyse these models and their impact on availability in detail in our guide to N, N+1 and 2N redundancy in the datacenter. The key point is that UPS redundancy must be matched by redundant distribution paths too: there is little value in duplicating the UPS units if all the power then passes through a single switchboard.
Maintenance: Batteries Are the Weak Point
Here is the most important lesson of the whole article: the electronics of a UPS are extraordinarily reliable, but batteries degrade. They lose capacity with every cycle and, above all, with temperature: it is estimated that a VRLA battery's lifespan halves for every 10 °C above its optimal temperature. A battery that on paper should deliver 10 minutes of autonomy may have dropped to 2 without anyone noticing, until the day of the blackout.
That is exactly why so many mains outages end in service downtime: not because the UPS was badly designed, but because its batteries were exhausted and nobody had checked. The UPS switched to battery as it should, but the battery did not last the seconds needed. So, more than installing a good UPS, what really matters is maintaining it and, above all, testing it.
A serious maintenance plan includes:
- check_circle Continuous, remote monitoring: voltage, current, temperature and the state of every battery string, with alarms that warn before a problem turns into a failure.
- check_circle Periodic discharge tests: subjecting the system to a controlled discharge to measure the batteries' real autonomy, not the theoretical one, and to detect degraded cells before they fail under pressure.
- check_circle Controlling battery-room temperature: keeping it in its optimal range is the cheapest way to extend battery life and avoid surprises.
- check_circle Planned replacement: replacing banks before they reach the end of their estimated life, without waiting for failure, and recording the date of each change.
Testing for real — including full transfers to generator — is uncomfortable and carries some operational risk, and that is precisely why many organisations avoid it. But a backup system that is never tested is, in reality, an assumption. The only way to know that the UPS-battery-generator chain will work is to have seen it work.
EasyDataHost: Redundant Online UPS and Generators
At EasyDataHost the technical room's power is designed with the same philosophy we have described in this article:
- arrow_right Online double-conversion UPS in a redundant configuration, so that the failure or maintenance of a single unit never interrupts the supply to hosted equipment.
- arrow_right Gensets that take over from the UPS for long-duration autonomy during an extended mains outage.
- arrow_right Continuous monitoring of battery health, loads and temperature, with maintenance plans and periodic tests to guarantee the backup responds when needed.
- arrow_right Redundant power distribution from the UPS all the way to the rack, so resilience is not lost in the final stretch.
All of this infrastructure runs in our own datacenter in Spain, with ISO 27001 certification and ENS compliance. You can learn more about our facilities and their electrical design, or contact our team if you need advice on the resilience of your infrastructure.
Frequently Asked Questions
What type of UPS is used in a datacenter?
The standard is the online double-conversion UPS. The load is always fed from the inverter, with zero transfer time during an outage and full isolation from mains disturbances. Offline and line-interactive topologies are reserved for non-critical equipment or SME environments.
How long do UPS batteries last?
VRLA lead-acid batteries typically last 3 to 5 years; lithium-ion, around 10 years and with better temperature tolerance. In both cases they degrade over time and with each cycle, so they must be monitored continuously and replaced before their real capacity drops below what is needed.
Why do you need a UPS if you already have generators?
The UPS covers the seconds between a mains outage and the generators starting and stabilising, an interval in which without a battery the load would lose power. It also provides permanent protection against micro-outages, surges, voltage sags and electrical noise that generators do not filter.
Conclusion
The UPS is the component that turns a mains outage into a mere hiccup rather than a service outage. Choosing the right topology, the correct battery and the right sizing is only half the job; the other half is maintaining and testing it:
- arrow_right The UPS is the critical bridge between a mains outage and the generators starting, and the permanent filter against micro-outages, surges and electrical noise.
- arrow_right Online double conversion is the datacenter standard for its zero transfer time and full isolation from the mains.
- arrow_right Lithium-ion is gaining ground on lead-acid for lifespan, space and thermal tolerance; the flywheel is a battery-free alternative to cover the generator start.
- arrow_right Size by real load with a growth margin, apply N+1 or 2N redundancy and, above all, monitor, test and replace the batteries: they are the weak point of the whole system.