TPC Benchmarks Now Measure Server Power Use

By Andy Patrizio (Send Email)
Posted Feb 3, 2010


The Transaction Processing Performance Council (TPC) today announced its newest specification, TPC-Energy, which measures the energy usage of a server to gauge not only transactions per second but transactions per watt as well. The benchmarking standards group has added per-watt rating to how it measures server transactions.

TPC benchmarks are a standard measure of performance used by every major server vendor to boast about the speed of their hardware. The top performers list is constantly changing as vendors release new servers.

But the one thing those lists didn't discuss was how much energy was burned to achieve those results. Two servers may come within a whisker of each other in terms of performance numbers, but if one vendor needed five times the hardware and consumed five times as much power to get similar results as the other, a buyer might want to know that detail.

Plus, it's needed to keep the vendors honest. Vendors can be a little sneaky some times by listing the power draw of a bare bones system that goes up exponentially once you add all the pieces you need to actually do the job,.

The TPC-Energy specification provides energy usage measurements that a typical customer of a particular benchmarked system would consume. It measures the power of the whole system running a typical end-user benchmark. It provides energy metrics that are comparable between systems for a particular TPC benchmark.

Most important, it's not a stand-alone benchmark. It works concurrently with the TPC-C benchmark, which measures transactions per second. So not only can you measure transactions per second, you also know transactions per watt of the configuration.

That's becoming very important to customers, according to Mike Nikolaiev, chairman of the TPC-Energy Specification Committee. With many datacenters maxed out, CIOs are asking how much more performance power can be squeezed into their current power environment, because their power envelop isn't getting any bigger.

"What we're hearing from the industry is performance and pricing have always been at the top of most vendors' lists in terms of buying decisions, but now we're hearing energy consumption is right up there at the top. It used to be at the bottom of the list," he told InternetNews.com.

"We want to provide a level playing field and full disclosure for customers to decide if [a server] is close to what they want or if it's something they would never implement in their datacenter," Nikolaiev added.

Twenty-two of the 25 members of the TPC committee voted on the final specification and it received a 100 percent 'yes' vote, something that has not happened before, Nikolaiev said. Results will be posted on the TPC benchmark pages in the next 30 to 60 days.

"Everyone felt it provides an even playing field, that they could all participate in and not be disadvantaged," he said.

More detailed server benchmarks

Michelle Bailey, research vice president for the Enterprise Platforms and Datacenter Trends division at IDC, said the industry has been in need of some kind of measurement like this for a while.

"I would say that the market is looking much more for what the TPC is providing than what is out there today," she told InternetNews.com. "PUE only tells part of the story. What you want to look at is how much work is getting done, and they've been really good at measuring that."

PUE, or Power Usage Effectiveness, is the measurement of how much power is used in a datacenter overall to power the systems and to cool them. It is the cumulative of both system power and cooling power wattage.

The Uptime Institute estimates the average PUE in datacenters is 2.5, meaning for every watt of power use to power the systems, 1.5 watts is needed to cool them. Microsoft (NASDAQ: MSFT) and Google (NASDAQW: GOOG) have had an arms race in their datacenters to get as low as possible. Google's lowest is 1.19, while Microsoft is aiming for 1.16.

"The market today is looking at PUE, which is total energy drawn divided by the IT load. That really tells a small part of the story and doesn't give you any visibility into that application and how much automation that app is supporting. Really, the market needs something that looks at the workloads of energy," said Bailey

Andy Patrizio is a senior editor at InternetNews.com, the news service of Internet.com, the network for technology professionals.

Follow ServerWatch on Twitter

Page 1 of 1


Comment and Contribute

Your name/nickname

Your email

(Maximum characters: 1200). You have characters left.