Five key elements to selecting a PDU

The National Grid has recently reported that there is just 1.2% spare capacity. Any additional energy demands, something as simple as a cold winter, could see an increase in the mains disturbances such as brown outs and black outs which will have a significant impact on the up time of our data centre infrastructure.

There has been a great deal written about reducing energy consumption using new IT cooling methods such as in-line cooling and aisle containment systems which, is where the majority of electrical energy consumption is consumed. But electrical power distribution is also an area that can be optimised, and that requires a review of the internal power distribution of a rack, i.e. the Power Distribution Unit (PDU).

There are five key elements to selecting PDU: 

1. Does it do the job?

PDUs distribute power, but to match your requirements to the right PDU you should consider: 

• What is the overall current consumption required from the PDU? This is dependent on the aggregated power consumption of the IT devices that are connected to the PDU. PDUs are rated in amps, such as 3.16 and 32 amps, so match the PDU to the nearest rating.
• Do I need single or three phase power? Again this is dependent on the total demand of the IT devices; as IT power demand increases it may be necessary for a three phase PDU to be used, typically where demand exceeds 8kW per rack.
• How many sockets and what type of sockets do I need? The number of sockets is dependent on the number of devices you are using, e.g. ten servers would need ten sockets. The socket type is typically determined by the device that is using that socket. Commonly used internationally are C13 outlets rated to eight amps or C19 outlets rated to 16A or, more often than not, a mixture of both.
• Do I need intelligent PDU metering and switching capability? Advance features such as remote monitoring and switching sockets off remotely really can change your understanding and control of your power consumption. They also improve resilience.

2. Does it do it efficiently?
To reduce power we must monitor it and then put in place strategies in response.

Monitoring can done either at the incomer of the PDU in terms of the total power being drawn from the whole PDU, or it can be focused on an individual device’s demands from the socket. The latter provides a greater granularity for analysis but not all meters have the same accuracy, it may range from one to five percent, so it is worth investigating into the technical specifications.

PDU electronics also consume power at levels which can be significant, especially in a large scale data centre, given that it can range from as little as six to 60 watts. If intelligent PDUs are installed in a 200 rack data centre with dual PDUs in the rack then their power consumption could be as little as 2.4kW or as much as 24kW.

The switching function allows users to remotely shutdown individual sockets and to ‘hard boot’ any device that is connected to it. It is worth checking which type of switch device (or relay) is within the PDU construction, either monostable or bi-stable. The monostable needs a permanent supply of power to change state and stay there. Bringing it back means removing the power. 

The bi-stable needs to be pulsed with electricity to change state and pulsed again to come back. Bi-stable therefore uses less power than monostable, which can again be significant in terms of the aggregate consumption from a lot of sockets.

3. How reliable is it?
Typically bi-stable relays have a higher Mean Time Between Failure (MTBF) than the lower-cost monostable option, in excess of 80,000 hours. This is significant given that most PDUs operate at the back of the rack where ambient temperatures can exceed 40°C, which, unsurprisingly, can shorten the lifespan of the conventional relay. 

Metering also an impact on the overall resilience of the power system. It allows users to determine their power consumption and how close the system may be getting to the tripping points of the circuit breakers, particularly if warning alarms are also enabled. 

4. Is it easy to integrate (mechanically and through software)?
Larger PDUs must fit into the back of the rack, however if possible they should not extend into the 19” space because of the potential this has for conflict with 19” mountable equipment. So mounting PDU positions that sit in parallel with the 19” angles can be beneficial and should be possible even in 600 wide racks if the PDU profile is slim enough. 

Most intelligent PDUs use SNMP protocol to communicate so simple integration into a DCIM or NMS is easily achieved.

5. Will it last the life time of the data centre?
An almost impossible question to answer because it is very hard to predict the development of future data centre technologies! It would be sensible to allow for future capacity by providing additional sockets within the PDU and also to ensure that you have spare capacity from the PDU incoming feed. 

If you work in a fast-changing environment then it is worth considering modular PDUs that allow you to add additional sockets if required. They are slightly more expensive but they offer greater flexibility enabling you to change the socket configuration without taking the IT system down.

Summary
The constraints of the electrical grid and the need to lower our power consumption are two issues that are not going away and we must all respond accordingly. PDUs allow you to profile your IT power demands, and if you can’t measure it you cannot improve it.