US gov polishes Energy Stars for servers
Here is perhaps the best sign that computing has really gone mainstream: Starting today, servers can be equipped with the same kind of yellow Energy Star tags as a stove, microwave, or refrigerator.
After more than two years of development in conjunction with server makers, power companies, data center operators, and other interested parties, the U.S. Environmental Protection Agency today released the first Energy Star rating specification for servers.
On Friday, the server makers participating in the EPA effort, which will put Energy Star labels on servers that meet energy efficiency and environmental protection standards, received the detailed specs that allow them to certify their machines and receive a rating. The 1.0 spec for the rating is going to cause a lot of grumbling, as any standard in the server racket does. The first pass of the Energy Star 1.0 spec for servers only includes rack and tower servers with no more than four processor sockets and does not cover blade servers (including the blades and their chasses), fault tolerant servers, server appliances, multi-node servers (such as preconfigured clusters), storage equipment, or networking equipment.
According to the document sent to the server makers, EPA hopes to add blades, multi-node, fault tolerant, and server appliances in the next phase, which is supposed to be done by October 15. At that time, it hopes to have support in the Energy Star rating spec for machines that have more than four processor sockets. EPA says that it will investigate giving storage and networking equipment their own Energy Star ratings under future and separate specifications.
The Energy Star 1.0 specification for servers - which EPA was good enough to send to El Reg - includes very precise rules about the components in the servers and how efficiently they run. For instance, it stipulates what power supplies can be used and at what efficiency these power supplies have to be running at for specific loads. It also stipulates what the idle power consumption of specific single-socket or dual-socket servers can be, depending on whether or not they have a service processor in them.
So, a single-socket box with no service processor (what EPA calls a standard server) has to have an idle power draw of 55 watts, while a single-socket box with a service processor (what EPA calls a managed server) has to be 65 watts or lower. A two-socket standard server has to consume 100 watts or less when it is idling, and a two-socket box can be no more than 150 watts. The assumption is that the servers all have one disk drive in them, and there is a lack of clarity about how many processors are allowed in the two-socket boxes to give the idle power wattage. (It had better be two sockets with two processors with whatever number of cores in the configuration).
The EPA is allowing 20 watts of power draw for redundant power supplies in a server, 8 watts per disk drive above the base drive, 2 watts per gigabyte of memory above the base 4 GB allowed, and 2 watts for any additional Gigabit Ethernet ports. (A 10 Gigabit Ethernet port is rated at 8 watts). As far as the spec is concerned, a disk drive is the same as a solid state drive, although they have radically different power draws and I/O performance.
The agency is warning that machines that have more than two processor sockets will have to have processor-level power governing technologies to earn the Energy Star label and that system makers will have to enable these features by default in their server BIOSes to get the certification. This will not be required on single- and dual-socket servers. All servers that have service processors and any server that has more than two sockets is also expected to have features to quantify the wall power that the systems draw as they are running.
The EPA wants to use this number to see if a machine meets the expected idle power ratings. How this data gets collected - whether by the microcode the service processor or by the operating system running on the machine - is up to the server maker. But EPA doesn't want to have to run around with a power meter testing configurations.
The grumbling was instantaneous upon the release of the documents that the Energy Star rating for servers was based on idle power for the machines, not under a particular benchmark load. But the EPA knows it needs to do this and says that it is working to get multiple workloads into the next release of the spec.
It seems very likely that the workload used by the EPA will be none other than the SPECpower_ssj2008 server power benchmark that the EPA and various server makers have created in conjunction with the Standard Performance Evaluation Corporation.
The SPEC power benchmark launched in December 2007 and updated with a 1.1 specification of its own in April of this year that added blade servers (which don't have their own power supplies, but rather share them collectively within a single chassis). The SPEC power benchmark was the default benchmark in the sample form that EPA expects server makers to fill out to get their Energy Star rating.
The exact metric of the Energy Star rating was not clear, but it clearly should be something simple, like SPEC power_ssj2008 operations per second per watt. A simple number for a precise configuration.
This is not just a tree-hugging effort on the part of the EPA. The Energy Star rating system used with household appliances and PCs has, in fact, compelled people to make intelligent choices about the electronic gear they acquire. Plus, the U.S. government in many cases stipulates that its agencies buy gear with specific Energy Star ratings. If Uncle Sam puts Energy Star ratings into its request for proposal process, you can bet that server makers are going to start jumping through hoops to get their gear certified.
But there is a larger goal. According to the EPA's calculations, the machines that earn the Energy Star label will be approximately 30 percent more energy efficient than those that don't, and if all servers sold in the U.S. were designed to meet the specification, energy savings could reach $800m a year and enough greenhouse gases could be removed from the air to compensate for the carbon dioxide pumped out by 1 million cars.
Return to green news headlines
View Green News Archive