Monitoring and managing power consumption

Reg Readers tell it like it really is

Trevor Pott
Infrastructure Support Engineer

Power consumption and cooling are two issues that cannot be disentangled. Eventually you will put too many servers on a single breaker, or notice that servers in the back closet don’t work anymore if you close the door. Once a company has been faced with this realisation, they have begun the long road towards datacentre design.

Every watt consumed by a server is a watt you have to figure out how to cool. For the small business with a single server under a desk, neither cooling nor power consumption are likely to be an immediate issue. When you are small, power and cooling can be as simple as a second breaker and leaving the door open. When you are Google, both factors are such important considerations that they determine the location and specific design of billion dollar data centres. Somewhere in between is the real world of everyday datacentre operations.

Simple tools like a kill-a-watt or a power distribution unit (PDU) with a built-in ammeter can tell you what the power draw of your servers are. Test the servers under heavy stress, as modern servers are fairly good at backing down their power consumption when they are idle. You need to get an idea of what they are pulling when loaded. Once you know how much power the servers consume you can start doing a little basic maths to see how much power you will need for future expansion. Factor in that you have to cool all of that heat, and that the chillers will cost you power as well.

In the long run, you will need to measure a server’s power draw over time. Any decent uninterruptible power supply (UPS) should be able to give you information on how much load is being drawn from it. UPSs are often equipped to send statistics on power usage to a central monitoring server from which you can collect information for later analysis. Like UPSs, the nicer PDUs are networked and are granular enough to allow per-socket monitoring.

Monitoring of the chillers duty cycles can help give you an idea of how much headroom you have to add servers to the datacentre. If the chillers are fully engaged for the entirety of your datacentre’s peak period, it’s probably not a good idea to add servers without adding chillers. Another good idea is to invest in thermometers that can record statistics. You can get them as network-attached devices, and they are fantastic at helping you find hot spots in your datacentre.

Proper power planning goes beyond simply ensuring that you have enough breakers pulled into your datacentre. It also means ensuring that you aren’t overtaxing your UPSs. Monitor your usage, plan for peak consumption, and above all leave yourself headroom for growth.