Several top computer industry companies are on the road to creating a universal set of measurements for server power consumption—something they hope will become the equivalent of a server miles per gallon—in an effort to help businesses grapple with rising data center energy bills.
The group, dubbed Eco Forum, includes representatives from Advanced Micro Devices, Google, Intel, Microsoft and Sun Microsystems, as well as the U.S. Department of Energys EnergyStar program.
It is still in the early stages of its work—it has met once thus far on March 27 at Suns corporate campus. However, the group has already started laying the foundations for what it hopes will be broadly acceptable server power consumption measurements, which could be used to determine how much electricity a given machine uses for doing a specific task, allowing similar machines to be rated against each other.
The forums work comes as concerns about server energy consumption are growing among senior IT managers, due to rising electric rates, higher oil prices and hefty increases in new server deployments over the past few years.
Yet, at the moment, there are no universally accepted tests—like the fuel mileage ratings in the auto industry—available for those managers to use in comparing different machines and making purchasing decisions about them.
If successful, the groups proposed measurements could help create the first broadly accepted server power miles per gallon ratings.
“The consensus among [the group] is that power is a concern—that customers are seeing it, manufacturers are seeing it, and it is something that were all going to need to deal with,” said Ed Hunter, director of Sun Microsystems Eco-Responsibility Initiative and one of the Eco Forums masters of ceremonies, in Santa Clara, Calif.
However, “Theres [also] a consensus that there isnt a good way for customers to really compare the energy efficiency of different products that they might buy. So what were going to try is to drive some measurements for how they might do that.”
To that end, the group is defining the set of procedures that people could follow to determine a servers energy consumption—therefore revealing information about its relative efficiency—during a given task.
The procedures to do this will have to be made as specific as possible, defining everything from what performance benchmarks to use to things like the locations of temperature sensors on server racks, to produce results IT managers can use for comparative purposes, Hunter said.
Ideally, the tools would measure the amount of electricity a machine draws at the wall or electrical socket, allowing them to account for its processors, memory and all additional components, including hard drives and power supplies, over different workloads, group members have said.
But given that different companies have different needs—for example, some rely heavily on databases, while others focus on Web transactions—and servers are configured differently to meet them, the group has said it will aim to create a suite of metrics.
Next Page: Choosing relevant benchmarks.
Page 2
Thus, some of the groundwork the group is undertaking now includes choosing the types of benchmarks that customers will find most relevant and sorting out server hardware—both the types of servers the measurements should address and how to configure them. Other issues include identifying and specifying air temperature ranges and temperature probe placement, Hunter said.
At issue for the groups own relevance are ensuring that it selects benchmarks, servers and server configurations, and environmental parameters that are relevant to customers and difficult to tamper with.
Defining air temperatures, for one, would help prevent a tester from feeding 50-degree air into one machine and 100-degree air into another, which would affect test results.
The forum hopes to present the results of its initial work at a meeting scheduled to take place in June in Washington D.C. There, the group is also likely to up the ante by discussing issues such as the design of server power supplies, Hunter said.
“I think what youll see in the June forum [is the group saying], Heres a list of problems that we think are important to address, and then asking the industry … what they want—how they want these addressed.”
Aside from helping to compare individual machines, the forums metrics might also be used to project servers future power consumption, thereby helping companies create more accurate data center electricity budgets, members have said.
Hunter stopped well short of calling its measurements for server power an industry standard, however.
“At this point no one is talking about any energy standards,” he said. “[One thing] I would be wary of is us starting off saying, OK, were going to create the standard.”
Given that many manufacturers involved—Sun, for one, created SWaP (Space, Watts and Performance) to take stock of data center efficiency—have already made attempts to measure their servers power consumption or assist their customers in controlling data center power, trying to force its work on the industry as a standard might risk delaying its progress, he said.
Indeed, some of the forums efforts are also likely to go into garnering broad-based support. Some of that may come soon as AMD is expected to announce in the near future a parallel effort to gather numerous industry players to focus on improving energy consumption in the data center.
Power “is something thats very important to people these days, and having a way to measure it might be a really good idea,” said Gordon Haff, analyst at Illuminata in Nashua, N.H., in a recent interview.
But Haff also cautioned that, because benchmarks can be skewed toward one type of result or another at times, the forum might also find challenges in gaining a critical mass of supporters.
Hence the focus on gaining a broad swath of industry input versus trying to force a standard, Hunter said.
“The purpose of the discussion [among the group] is to start to build the framework for how you go about taking these measurements, and [ensure] the metrics are customer-driven,” he said.
“What we wanted to avoid is getting into a metrics war.”