Imagine a situation in which a financial services company employee is working on a document when the computer system goes down due to a power failure. The criticality of the document defines the potential loss. Now extrapolate that scenario to one in which an entire data center power failure
And, were it to be established that the very data center power design was responsible for the total power failure, the management would be justified in blowing its top. Fact is, at the time of data center power design, organizations often overlook seemingly innocuous details that actually have significant bearing on the normal functioning of the data center. Let’s take a look at five common mistakes made when designing data center power infrastructure.
1. Incorrect sizing of data center power capacities
If one goes by the nameplate power ratings of servers, one could end up over-provisioning data center power capacities. Let us assume a server with a power rating of 1kVA. Even with peak workloads, the server CPU would be utilized at most 50%. For 1,000 such servers, the data center would apparently require a 1,000 kVA power supply system; as a caution, redundancy is built in by doubling or tripling the power capacity. Consequently, the infrastructure would end up being underutilized and hence inefficient.
The correct data center power design approach is to study historical power consumption trends to arrive at the required optimum. Alternatively, if one wishes to go with the nameplate power rating, the design should enable automatic switching on/off of the diesel backup generators to ensure dynamic capacity utilization within the optimum efficiency range.
2. Single point failure:
Organizations often end up connecting multiple components to a single power line, resulting in a single point of failure in the data center power system. For instance, if the power requirement is 1000 kVA and the UPS systems are implemented with inbuilt redundancy such that three 500 kVA UPS systems connect to a single power distribution unit (PDU), all the UPS systems would go down in case the PDU fails.
To avoid this, an alternative design could have two PDUs each connecting to two 500 kVA UPS units through individual cables in a dual bus architecture. Thus, if either a PDU or a UPS fails, the others would meet the overall requirement.
3. Choice of data center power equipment:
Power supply components in the data center power design could be incorrectly chosen. For instance, most current generation servers have a leading power factor value. This means that these servers draw a leading current – whereby for the base angle of current and voltage, the current leads the voltage value.
While a 500 kVA UPS with a 0.8 leading power factor will deliver 300 KW, a UPS with a 0.8 lagging power factor will deliver 400 KW. Any mismatch between power factors of the UPS and server will result in an overload, and trip the UPS.
4. Design of the automatic transfer mechanism
Power transfer between the utility power and the backup generator occurs through a transfer switch. Data center power design in India generally employs four-pole circuit breakers or power contactors for the transfer, which requires disconnecting the neutral terminal.
However, current generation UPS systems being transformer-less, require an input-neutral connection. When the generator kicks in, the UPS has no neutral terminal to refer to, leading to a floating neutral that further results in a floating load. This causes server freezing or tripping. The solution is to use an automatic transfer switch with an overlapping neutral mechanism.
5. Cabling between server and UPS:
Most of the IT load results in a single-phase power load. This generates a large amount of harmonic distortion, which can cause a voltage drop, excess heating and resultant overload. Most data center power design ends up under-sizing the cable connecting the server to the PDU. If a server requires a 100 sq mm cable, then one must put in a 130 sq mm cable in order to withstand the peak harmonic distortion levels and cable heating. The neutral current returned from the server to the UPS is twice that of the phase current. Therefore, the size of the neutral cable needs to be double that of the phase cable, to prevent the setup from exploding.
About the author: Venkat Rao is country manager for medium and large UPS systems at Emerson Network Power India. He has 14 years of experience in the fields of power quality solutions and power generation.
(As told to Harshal Kallyanpur)
This was first published in April 2011