How a Southeast Asia Data Center Cut Downtime with Online UPS

A colocation operator in Southeast Asia cut unplanned customer downtime by roughly 92 percent and held a 99.995 percent annual uptime SLA after replacing standby units with online double-conversion UPS at every rack row. The measured saving came to about 38,000 US dollars in avoided SLA penalties and recovered engineering time in the first twelve months, which paid back the upgrade inside nine months.

The Site and Its Original Power Setup

The facility serves enterprise and cloud tenants from a single building in a tropical coastal city where the grid is generally available but experiences frequent short breaks and wide voltage swings during storm season. At the start of the project the data hall relied on basic standby UPS units placed under individual racks, each feeding a small cluster of servers with no central coordination.

Those units did provide battery backup, but they transferred the load to raw utility power the moment mains returned, and their voltage correction was either absent or limited to a narrow window. For most of the year that was tolerable. During the wet season it became the dominant source of tenant complaints.

The Problem: Short Grid Interruptions and Voltage Swings

The operator logged the events for three months before committing to a change. The pattern was not long blackouts but dozens of sub-second to multi-second interruptions per week, plus voltage excursions that drifted outside the tolerance of tenant power supplies. Each event triggered a transfer on the standby units, and a fraction of those transfers produced a visible dip at the server inlet.

Because the load was sensitive storage and virtualisation hardware, even a brief disturbance forced a node to drop or a cluster to re-synchronise. Tenant-visible downtime accumulated to roughly two hundred minutes per year, most of it in short fragments that were hard to explain on a status page.

The Solution: Online Double-Conversion at the Row

The operator standardised on row-level online double-conversion UPS modules, one per rack row, configured as a shared N+1 pool rather than a box per rack. In this topology the inverter always supplies the load, so utility disturbances never reach the servers. Voltage and frequency are reconstructed independently of the incoming supply, and transfer time to battery is effectively zero.

Battery runtime was sized to cover only generator start and stabilisation, about eight minutes, with the existing diesel generators taking over for any longer event. Lithium iron phosphate strings were chosen for their tolerance of the warm plant-room ambient and their faster recharge between events.

Measured Results: Downtime, SLA and Cost

Twelve months after commissioning, the numbers were unambiguous. Customer-affecting downtime fell from roughly two hundred minutes per year to under fifteen, a reduction of about 92 percent. The hall held its 99.995 percent SLA through the storm season for the first time, and SLA penalty charges dropped to near zero.

Engineering time spent on power-related incident calls fell by an estimated sixty percent, which the operator valued at around 38,000 dollars annually once recovered hours were priced against project work. The full upgrade, including new modules, lithium batteries and re-cabling, paid back inside nine months on those two lines alone.

Why Online Topology Was the Deciding Factor

A line-interactive or standby unit would have reduced some of the symptoms but not the root cause. Those topologies still pass conditioned or raw mains to the load between transfers, so the voltage swings that triggered tenant incidents would have remained. Only continuous inversion removes the utility entirely from the failure path.

For a multi-tenant hall where one disturbance can take several customers offline at once, that isolation is worth the efficiency and capital premium. The operator judged the avoided penalty and reputation risk as a larger number than the extra conversion loss, and the measured result supported that judgement.

Lessons Other Operators Can Apply

• Log real events for a season before specifying; averages hide the short interruptions that actually hurt.
• Size runtime to your restoration plan, not to a generic "ten minutes", and let generators carry the long tail.
• Pool modules as N+1 at the row to cut spare count and simplify service.
• Choose battery chemistry for the room you have, not the datasheet you prefer.
• Treat transfer time to battery as a design requirement, not a footnote.

The row-level power architecture we deployed is detailed on our https://www.upsboss.com/data-center/ page, and practical commissioning questions are answered on the https://www.upsboss.com/faq/ page. If you run a similar hall and want the same event-driven approach, send us your incident log and we will size the upgrade against your real failure pattern.

Key takeaway: the operator cut downtime about 92 percent and held a 99.995 percent SLA by moving to row-level online double-conversion UPS with lithium batteries and generator-backed runtime, recovering the cost in nine months through avoided penalties and reclaimed engineering time.

Frequently Asked Questions

Why not just add more standby units to fix the downtime?
Standby units still feed the load from mains between transfers, so the voltage swings that caused the incidents would remain. Online topology removes the utility from the path entirely, which is what actually stopped the tenant-visible dips.

Was the extra efficiency loss worth it?
On this hall the avoided SLA penalties and engineering time, about 38,000 dollars a year, outweighed the conversion loss. For a multi-tenant site the reputation risk of one shared outage was the larger number.

Could a smaller runtime have worked?
Yes. With reliable generators the batteries only had to cover start and stabilisation, about eight minutes. Sizing to the restoration plan rather than a generic figure kept the battery footprint and cost down.