The recent release of Energy Star requirements for servers may have coaxed a collective sigh of relief from datacenter operators who've felt the pressure to cut energy waste. The well-known Energy Star symbol, after all, is associated with energy efficiency. Thus, having that familiar logo stamped on a server might suggest it will deliver equal or better performance than its non-Energy Star rivals while consuming less electricity.
Unfortunately, the new Energy Star requirements for servers have enough shortcomings that they are unsuitable to be primary criteria for the purchase of new hardware equipment. That's not to say they have no value whatsoever; as observed by Subodh Bapat, vice president of energy efficiency and distinguished engineer at Sun, they are "a good first step."
A bit of background: These new Energy Star server requirements for servers have been under development for about a year and a half. The EPA drew on feedback from stakeholders, such as server vendors (including Sun), to hammer out the final requirements released late last week. In a nutshell, the criteria were limited to machines with one to four sockets and at least one hard drive. Additionally, blades weren't included, though the EPA said requirements for those should be out in a couple of months.
Idle servers are the devil's tools
So what's wrong with the new Energy Star criteria? Perhaps the most significant drawback is that a qualifying server need only demonstrate energy efficiency when it's in idle -- that is, powered on but doing no work.
To better illustrate the problem, imagine you're at a used car lot where a shifty salesperson is trying to push you to buy an SUV. His selling point: "This baby uses the same amount of fuel as a hybrid sedan -- when you're not moving."
You probably wouldn't be too impressed by this fact, as you likely intend to spend more time driving your car when it's turned on, rather than idling at stop signs or in the driveway. Thus, you want to know how many miles per gallon the vehicle gets in the city and on the freeway.
Now back to the Energy Star server requirements: As I noted, they consider only how much electricity a server consumes when it's on but doing no work. Granted, underutilized servers abound in the datacenter, so server power consumption at idle is relevant. However, datacenter operators are becoming increasingly aware of the problem. Organizations are tracking down underutilized machines and employing virtualization to get as much performance per watt as possible from their hardware. Others are dabbling in powering down servers entirely when they're not in use. In short, weighing energy consumption at idle merely scratches the surface of a server's overall power usage.
[ Virtualization was a key tool for reducing energy consumption among 2009 InfoWorld Green 15 winners. ]
Ideally, the Energy Star spec for servers would consider server performance per watt (the equivalent of miles per gallon in cars) at various levels of server utilization: X processes per watt at 25 percent utilization, Y processes per watt at 50 percent utilization, Z processes per watt at 75 percent utilization, and so on.
Of course, that's easy for me to say. There continues to be much debate and, yes, controversy as to the best way to measure performance per watt for servers. SPEC took a stab at it in 2007 with its SPECpower benchmark, which goes much further than Energy Star and still garnered some criticism for being insufficient. Similarly, independent analysts such as Neal Nelson have devised benchmarks for measuring performance per watt -- but thus far, nothing has been fully embraced by the industry. Debate centers around what sort of workloads to measure, where server temperature should fit into the equation, where virtualization functionality fits into the picture, and so on.
Back, though, to Energy Star requirement for servers, Sun's Bapat pointed out another drawback to the program: It doesn't take into account how many cores per processor a machine has. "The fact is, when you go from a server that has four processors with two cores each to two processors with four cores each, you save energy. That's not recognized by the spec," he said. "If you're shipping a server with one processor, it doesn't matter if you have one core or two cores or four or eight. You still get the same idle power allowance. There's no benefit for the fact that you can do, say, eight times work with a fewer number of watts."
The IDG News Service reports yet another drawback to the Energy Star requirements: "Critics have also noted that resellers may reconfigure systems in the channel, by adding more memory and disk drives, for example. The EPA will rein in such behavior as much as it can, but customers should check that the configuration they receive is the one that qualified for the program in the first place."
Bapat wasn't entirely critical about the Energy Star program for servers. For example, a compliant server must be capable of measuring real-time power use, processor utilization, and air temperature, data that can help datacenter operators asses the overall efficiency of their operations. "Transparency is always a good thing. Energy Star requires the ability to report power consumption data pretty much across the range of utilization and at all times that the server is on. If you want to know how much [power is being consumed], you should be able to ask it and it should tell you. That's a very useful feature."
Additionally, the spec requires the use of more efficient power supplies, which generate less waste heat, thus requiring less cooling. It also requires compliant servers to come equipped with advanced power-management features to save energy across various operating states.
All in all, I agree with Bapat's assessment that Version 1.0 of the Energy Star requirements for servers is a good first step, but there's plenty more work to be done. In the meantime, datacenter operators will still have to do lots of homework before investing in new machines that fit their organization's data-crunching and energy-efficiency requirements.