Following the data center energy track

Energy conversion

An alternative way to look at information processing in a data center is to see it as an energy conversion machine. Energy that is converted in something useful: data. The energy efficiency of this conversion process is the ratio between the useful output and the input in energy terms. As we will see, using a simple model shows us that the conversion process is highly inefficient and therefore creating tremendous losses. Energy efficiencyPart of the problem (the inefficiency) is that a lot of people take the electrical power supply for information processing for granted. But information processing (thus data centers) are part of a complex electrical power value chain. People are mostly not aware of the energy losses in this value chain. Saving a kiloWatt at the end of this power value chain saves society a lot of energy and reduce carbon emission.

Model

The Power Loss model is composed out of three sub-systems:

  • Power sub-system: The efficiency of power plants varies widely with the technology used. In a traditional coal plant, only about 30-35% of the energy in the coal ends up as electricity on the other end of the generator. So called “supercritical” coal plants can reach efficiency levels in the mid-40’s, and the latest coal technology (gasification) is capable of efficiency levels above 60%. The most efficient gas-fired generators achieve a similar level of efficiency. After the electrical energy is generated, it has to be transported to the consumer by means of a power grid (transmission and distribution network). Generally speaking, t&d losses between 6% and 8% are considered normal. Obviously, there is a tremendous amount of energy left behind in the generation and t&d process.
  • DC infrastructure sub-system: Once electricity is supplied to a data center, various devices consume the electrical power. A data center has, from a power perspective, a supply chain that consists of four large building blocks: the IT infrastructure (servers, storage and network), the primary power supply (UPS, PDU, etc.), the secondary support supply (cooling, generator, air handling) and the tertiary support supply (lighting, office space).  Typically about 50% of total power usage in a data center is derived from cooling, power distribution, lighting, etc. The figures of the DC infrastructure power consumption are taken from APC.
  • IT infrastructure sub-system: Typically about fifty percent of total power usage in a data center is derived from the IT infrastructure. We have servers in all kind of flavors that have a different power consumption. Making use of  the report ”ESTIMATING TOTAL POWER CONSUMPTION BY SERVERS IN THE U.S. AND THE WORLD” of  Jonathan G. Koomey of Stanford University the power usage of low, mid and high range server  are estimated on 180, 420, and  4800 Watt. In the EPA Report to Congress on Server and Data Center Energy Efficiency is suggested that, servers will on average account for about 75 percent of total IT equipment energy use, storage devices will account for around 15 percent, and network equipment will account for around 10 percent. However, these percentages can vary greatly depending on the configuration, hosted applications, and data storage requirements of individual data centers. Further it isn’t unusual that on average just 10% of the processing capacity is actual used for information processing. At the end virtually all power consumed by the IT infrastructure is converted to heat.

Click on the model to make it readable.

Inefficiency

As you can see in the Power loss model in the first step, delivering power to the data center, we have already lost nearly 70% of the original energy provided by the original energy source. In the second step, delivering power to the IT infrastructure (servers, storage and network) we lose another 16% of the original energy source or 50% of the energy delivered at the doorstep of the data center. This leaves us with 13.9% of the original energy source to start the information processing and thus a loss of 86%. But even in the “information processing” we have severe losses. Most of the time the infrastructure is running idle but nevertheless is using energy. That brings us to a final 1.4% of the original energy to use for real information processing!

If we set up a data center case with 800 servers (600 low range, 200 mid range, 100 high range) we can see the stunning impact and loss of megawatts at each component of the power value chain.

So we must optimize the components of the power value chain? Of course but because of the specific properties of electrical power infrastructure, this infrastructure isn’t very flexible and adapt for change. So changes for optimization cost a lot of time (and money). If this is the case, shouldn’t we start to look more at the demand side of this power value chain then to the supply side? As the model shows saving 1 unit power consumption in information processing saves us about 98 units in the upstream of the power value chain.

A proper integrated architecture and design is needed to correct unnecessary redundancy and inappropriate use of infrastructure. This approach reduces not only the capital expenditure on server, network and storage systems but also the energy needed to run them.

As stated earlier the model gives you just a rough idea about power loss. There are several ways to enhance the model. But the first goal is to create awareness. You don’t have to wait to cut energy usage, you can start now by improving your IT design and right size your infrastructure capacity.

Bookmark and Share

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s