Today I’ll talk about how to conduct load testing in a data center.
I. The essence of dummy load testing is end-to-end stress verification.
Many people mistakenly believe that dummy load testing simply involves applying a load to the power supply, which is a fatal misconception. The real core is “simulating the end-to-end collaborative capability under real business scenarios”—from mains power access to UPS conversion, from diesel generator backup to cooling response, from power distribution to the server rack end ; if any link fails, the entire system may be paralyzed.
1. It’s not about testing equipment, it’s about testing the system.
In a certain data center project, the UPS performed normally when tested individually, but frequently tripped during joint testing. It was later discovered that the excessive distortion rate of the diesel generator’s output voltage waveform caused the UPS rectifier to malfunction. This illustrates the pitfalls of isolated testing: the compliance of a single device does not guarantee system compatibility; the coordinated response of the “mains power-UPS-diesel generator-power distribution-cooling” system must be tested.
2. It’s not a static measurement, but a dynamic one.
There is never a constant load in business operations—server startup and shutdown, virtual machine migration, and sudden traffic surges all cause load fluctuations. Last year, during testing at a cloud computing center, acceptance was signed off after only testing at 100% steady-state load. After going live, a sudden increase in load from 30% to 80% caused a power distribution switch in one of the server racks to trip. Real testing must include scenarios of “load step” and “sudden load increase and decrease” to verify the system’s dynamic response speed; this is the core of closely reflecting real business operations.
3. It’s not about measuring compliance, it’s about measuring the limits.
A 100% design capacity does not guarantee it can withstand a 100% load—in one project, a UPS was rated for 100% load capacity, but during testing, uneven current distribution occurred between modules at 80% load. A qualified test must include “extreme testing”: running the system at 110%-120% load for 1-2 hours to observe the system’s “critical failure point” and verify the reliable operation of overload protection. This is not “tampering with the equipment,” but rather providing sufficient “safety redundancy” for maintenance.
II. Equipment Selection: Don’t be fooled by parameter lists; scene matching is key.
There’s a wide variety of fake load testing equipment on the market, ranging in price from tens of thousands to millions of dollars. However, the key to choosing the right one isn’t how impressive the specifications are, but whether it matches your data center scenario. Many projects have suffered from blindly selecting equipment, spending a lot of money but failing to detect real problems.
1. Purely resistive load bank: Only suitable for basic power distribution testing; do not use it to test UPS systems.
Principle: Resistance heating consumes active power, and the power factor is fixed at 1.0.
What it can do: measure the current carrying capacity of power distribution circuits, the breaking capacity of switches, and the voltage drop of cables—for example, verifying whether the 1200A busbar of a certain cabinet can stably carry a 1000A load.
Things you absolutely must not do: test UPS efficiency, diesel generator load capacity, or reactive power compensation effect.
A painful lesson learned: A project used a purely resistive load to test UPS efficiency, displaying 95%. However, after going live, because the IT load contained a 30% inductive component, the UPS efficiency was only 91%, resulting in an additional 800,000 yuan in annual electricity costs. The reason is simple: a purely resistive load cannot simulate the inductive reactive power of a server, and therefore cannot measure the true losses of the UPS rectifier.