As energy continues to become more expensive and as global citizens begin to take notice of our collective impact on the environment, there are significant financial and social benefits for companies that undertake green initiatives. In this post, I provide a general overview of how server power supplies operate from an efficiency perspective and explain some ways more efficient power supplies and usage patterns are beneficial.
Power supply load and efficiency
A power supply is used to convert wall/line AC power to DC power, which can be distributed and used to power various components in a server, such as the motherboard and the hard drives. This conversion process has some overhead, which comes in the form of heat, and which reduces the overall efficiency of the power supply. The more efficient the power supply, the less heat that is generated. In simplistic terms, the efficiency of the power supply is the amount of energy that makes its way to the components needing the power. The remainder of the energy is given off as heat.
Don't forget that a power supply has to simply function, so there is a baseline operational cost that has to be factored into general usage. This comes right off the top of the efficiency metric.
Many single-use servers are overbuilt to handle potential demand peaks. This means that these servers often run at well below their specs and never make their way into the higher efficiency ranges on the power supply due to the fact that the power supplies are never pushed to their limits. When you add redundant power supplies to the mix, the efficiency view gets even worse. Remember, lower power supply unit (PSU) utilization equals lower PSU efficiency, so spreading the load among two PSUs creates a situation in which the PSUs, on average, operate at lower usage levels. There are compelling reasons for organizations to consider redundant PSUs in mission critical servers, but understanding the potential energy efficiency tradeoff is important when considering green initiatives.Now consider this: Are you running servers with very high wattage power supplies but not running them at an efficient level? If you aren't sure about what you're actually operating, you can compare the PSU rating in your server with what you're actually seeing by using a device such as a Kill A Watt power usage meter. Does that 1000 Watt power supply in your server actually draw between 400 and 600 Watts based on its hardware configuration and usage profile? If not, your server isn't running in the middle of the power curve, which is the sweet spot. (Note: Just because your PSU is rated for 1000W doesn't mean that it always draws 1000 Watts; it pulls only the electricity necessary to operate the system, which may be 400W to 600W.)
Server power supplies
Quality server power supplies tend to remain efficient even at high load and don't always drop off in a massive way. Recent shifts to technologies such as virtualization can have a significant impact on individual server power performance. Servers commissioned to be virtual hosts tend to be spec'd with more hardware that is pushed quite a lot harder than the hardware in single-use servers; as such, these hosts are much more likely to enter the PSU efficiency sweet spot. When you add to this the ability to reduce the number of servers using power, you can see that the individual activities start to add up to real energy savings. Further, newer power supplies are vastly more efficient than their years-old counterparts, often reaching efficiency levels far exceeding 90%.
A typical server power supply efficiency patternThe graph in Figure A depicts a typical server power supply efficiency pattern, omitting the efficiency percentages, which can vary by power supply. As you can see in Figure A, at below 40% to 50% utilization, power supplies aren't fully efficient. In fact, at a load of 20%, power supply efficiency can be as much as 10% lower than efficiency at the 50% load mark and even less efficient at lower levels. From about 60% usage to 100% usage, there is a noticeable efficiency drop-off as well, although this drop-off appears to be much less noticeable for server-based power supplies. In my research for this post, I saw some desktop power supplies that, at 100% usage, would drop to very poor efficiency levels, so there is a vast variety of operating reality out there. In other words, one size does not fit all, so your results may vary. Figure A
A typical server power supply efficiency pattern
Real-world power supply tests
Many of the real-world power supply tests were conducted by a third-party independent testing firm on behalf of the 80 PLUS® Program, a certification program intended to promote highly-efficient (greater than 80%) power supplies in servers. You can find the test results for many vendors at the test site. Although some of the results are out of date, you will see in many cases a similar usage pattern as the one in Figure A.
PSU Efficiency Savings Calculator
When it comes to measuring whether there is a return on investment, you should check out a PSU Efficiency Savings Calculator, which will show you the difference in efficiency of two power supplies and tell you how much you will save on an annual basis. (Before you use the calculator, you should have an idea about how much of a load you'll be placing on the power supply.) In my example, I would save $20 per year by moving from an 85% efficient PSU to a 90% efficient PSU. If the cost to buy the 90% efficient PSU was only $30 more than the 85% PSU, and the expected life of the server is three years, there is a positive ROI so I would go ahead and get the more efficient unit. If you multiply this across hundreds of servers, there is real money and energy being saved.
There are second order benefits to more efficient power supplies, too. Remember, the "wasted" energy in power supplies is given off as heat. Data center operators spend a lot of time and money working hard to keep rooms cool in order to ensure that equipment doesn't get too hot. As these organizations are spending untold amounts of money cooling the air in the data center, inefficient power supplies are adding to that workload. By reducing the heat load, there can be additional indirect energy savings since the cooling need can be reduced.
The Facebook way
Facebook has undertaken a massive effort to improve the power efficiency in its data centers. With a traditional power distribution configuration, Facebook engineers determined that the total loss of energy was 11% to 17% up to each server. By moving to a DC-based data center power distribution system and making a number of other changes, this loss is reduced to 2% up to the server.
The folks over at our sister site ZDNet recently posted a write-up about Facebook open sourcing its hardware design for its data centers, which includes information about this power savings.
Although the numbers can be small — $30 here, $10 there — the potential savings when it comes to energy usage can be incredible. One way to start is to carefully consider decisions about power supply and make sure that the workloads match the power capacity of the server in order to attain the best possible measures. When companies can tie an altruistic need to the bottom line, wonderful things can happen.
Since 1994, Scott Lowe has been providing technology solutions to a variety of organizations. After spending 10 years in multiple CIO roles, Scott is now an independent consultant, blogger, author, owner of The 1610 Group, and a Senior IT Executive with CampusWorks, Inc. Scott is available for consulting, writing, and speaking engagements and can be reached at email@example.com.