One of the great frustrations for many IT managers, especially those concerned with power consumption, is the inability to know beforehand how much power a new piece of hardware will draw. Most manufacturers simply don't publish this data. In the case of PC, server, and system vendors, there is arguably some justification for this policy, as each configuration is different, and so no fixed usage number would be a true reflection. A few hardware vendors provide basic power calculators for some PC and server models (here is
a calculator for PCs from Dell), but rarely do these provide the comprehensive information you need for an informed decision. The absence of data from vendors of components, where configuration is far less of a factor, is an equal pain and one that I expect will improve as reductions in power consumption increasingly become a competitive factor.
The upshot is that many IT managers have no real feel for how much power systems and components actually consume. For example, how much do all those desktops running late at night cost? How about workgroup servers? Blades vs. pizza-box servers? High-end PCs vs workstations? And so on.
One way to start getting a handle on this is to use a simple wattmeter and begin walking around the enterprise testing various configurations. The results, I assure you, are eye-opening. For starters, pick up an inexpensive meter such as the
Kill-A-Watt Electricity Usage Meter from P3 International.
As you can see, the meter has a three-prong outlet into which you plug the device you want to test, and then you choose the metric you're trying to quantify from the front row of buttons: volts, amps, watts, watt-hours (for long-lasting tests), and so on. The data is presented on a large four-digit display from which you will begin to get some sense of what's eating up your power budget.
Let's begin with desktops. Last year, our firm purchased a Dell Dimension E521 PC with a dual-core AMD processor, 2GB RAM, two hard-disk drives, and a mid-range graphics card that can drive two large flat panel monitors. (Today, this machine, called the Dell Inspiron E531, retails in this configuration for under $700.) When plugged in and running, but performing no particular task, the system consumes 110 watts, according to the watt meter.
This machine replaced a dual-Xeon Dell workstation with 3GB of RAM, two disks, and a lesser graphics card. That system, per the meter, used 191 watts at rest. But that difference is only part of the story. We ran benchmarks on both systems, using the
Sandra benchmark software and found a wide difference in platform capabilities. The tests rated the new system at 29.4 (where higher is better), while the old system came in at 22.7. Combining the two data points revealed that the old system delivered less than half performance per watt of the new PC. How then to translate the performance boost into further energy savings?
Maximizing That Benefit
Looking only at the power savings, the gains are modest: 1.9KWh per day. If your electrical utility charges $0.10/KWh, this number translates into savings of roughly $70/year. Multiplied by each desktop, however, the sums start to add up.
The economics become better when you include performance in the mix. One PC can do the work of two. If you were to use virtualization, you could run two PC instances on the new system-and the savings would now be considerably more. You would have the previous savings plus the entire power consumption of the second PC. In our case, if we saved the full consumption of the old Dell workstation, our new system would reduce daily consumption by 6.5KWKh and save us $238 per year. At this rate, we would have pay back on the new system in less than three years
on power savings alone.
On the surface, the math might seem incorrect, because two virtualized PCs on a single system would generate twice as much activity, so using the savings numbers calculated from a PC at rest are not reflective of the real world. This is a valid point, but it obscures another factor. Most PCs use very little of their computing power, even when they are so-called "busy." This is particularly true on dual-core or multiprocessor systems, where many applications can only run on one core; so running two applications at the same time presents little additional performance load, although admittedly it does consume more power.
The comparatively little stress applications place on PCs is meaningful in assessing PC (and server) consolidation. Reports from enterprises that employ virtualization for consolidation consistently point to an unexpected pattern: they are generally able to consolidate more systems on to one host than they originally expected-again because most PCs and many servers spend much of their time idle. In the example I am developing, the reality is that you probably could consolidate three or four PC loads onto the new system-and now the ROI is very quick and the savings become important.
None of this would be visible to a manager, however, without the ability to measure power consumption. And for this you need a tool, even as primitive as a simple watt meter to start to learn basic energy-consumption patterns.
As I will discuss in future columns, the ability to track consumption at fairly low levels is an important part of being able to intelligently plan the purchase and deployment of hardware. In the words attributed in multiple variations to Lord Kelvin: "You can't manage what you don't measure."
Andrew Binstock is the technology editor at GreenerComputing.com. His blog on software and technical matters can be found at http://binstock.blogspot.com.
