Deciding on a final power management strategy for server rooms can be a difficult task. Priorities must be assessed and service level requirements taken in to account. How long should systems be kept running once main power has been lost? Can some non-essential systems be shut down early to give essential services a greater runtime? Are there auxiliary devices such as routers and modems that also need to be covered--how about cooling systems? Something that should not be ignored is the possibility of an increase of requirements in the future.
Once questions like these have been pondered, we can start to calculate the requirements. As mentioned above, cooling is one area that will have a large impact on final power requirements. APC suggest that cooling can represent up to 50 percent of the data centres' requirements. In smaller environments, cooling may not be required while running on battery power; if, for example, a small server room would be sustained on backup power for 20 minutes before initiating its shutdown sequence, the room temperature may not rise above acceptable levels during those 20-30 minutes without cooling. Don’t forget to take in to account the time required to cool the room once power has been restored. Quite a few UPS units (such as those from APC) now include environmental monitoring probes as standard, which track temperature and humidity levels. This can allow the UPS to provide protection against not only power outages, but also air conditioning failure.
Calculating the maximum load a UPS will be required to carry is pretty simple. Hardware manufacturers usually publish figures stating the minimum and maximum requirements of their products. IBM provides server power consumption figures in its user guides (this example is for an x345) and HP provide a very nice power calculator spreadsheet for their Proliant range of servers. The figures we are most interested in for selecting a UPS are the input kilovolt-amperes (kVA).
If a device doesn’t give its power rating in kVA (or VA), then it can be calculated pretty easily:
(volt-amperes) VA = V (voltage) x A (maximum amperage)
(kilovolt-amperes) kVA = VA (volt-amperes) / 1000
You may notice variation between loads that you have calculated and the maximum loads quoted by manufacturers. Research has suggested that nameplate ratings are usually at least 33 percent in excess of the actual running load--this may be to allow for high start-up loads and compliance with regulatory bodies.
To avoid the expense and inconvenience of needing to revisit power requirements each time that additional hardware is brought into the environment, it’s a good idea to plan for any future expansion up front. If there is not information available on future plans, then it will probably be a good idea to work out your average power requirements per U of rack space. This way, we can have an idea of the general requirements that could be generated by the addition of new equipment to existing racks, or even the deployment of a completely new rackspace.
This has been a brief take on calculating power requirements for server rooms and small data centres--a nice whitepaper on the topic is available on TechRepublic published by APC. It’s worth a read.
I’d be interested to hear what type of setups readers are using to protect their power supplies, supply sizes, UPS ratings, systems coverage, and generators?