Issue 13, Dec 10/Jan 11
FOCUS METRICS
efficiency of their global infrastructure or [are] thinking about relocating data center resources across borders.” Most members of The Green Grid are multinationals, he points out.
Even with widespread global acceptance of PUE, Mathew acknowledges the metric will still only be fit to use for “preliminary” assessments of energy efficiency because there will always be site-specific characteristics that make a given data center unique. That is why work is ongoing on multiple other metrics, each meant to focus on one aspect of the data center’s performance instead of measuring its overall efficiency.
“We’re not coming up with a method that will give you a very solid apples-to-apples comparison,” he says. “Some people get hung up on the word ‘comparing.’”
Monroe, who had been involved in the efforts to define PUE and some of the follow-up metrics during his time as a member of The Green Grid, says PUE was meant as a stopgap measure to give us something to start with.
ONE PATH CHOSEN TO DIFFERING GOALS
While Mathew has observed some differences in perspective of the matter from country to country, they are usually marginal. One difference of opinion he noted was whether or not on-site generation or energy used to chill water should be included in the PUE calculation.
An example of a difference where geographical location plays a role is whether or not source- energy factors should be considered. Source- energy factor is the amount of energy it takes to generate and deliver one unit of energy to the end user. This factor varies from place to place. Another geography-specific variable is the difference in efficiency when burning natural gas at different altitudes, Azevedo adds.
According to Azevedo, differences in the levels of concern about energy efficiency from one part of the world to the other do not get in the way of the harmonization effort.
One would expect that British participants in the harmonization effort would be more preoccupied with metrics like the newly introduced Carbon Usage Effectiveness metric than others, given the UK’s CRC Energy Efficiency Scheme, but Azevedo says that is not the case.
While the Japanese industry is not as driven by a limited supply of power, data center energy efficiency is a priority because reduction in carbon emissions is a national goal, says Zen Kishimoto, principal analyst at the research firm AltaTerra. “In Japan, they don’t really have a power-shortage problem,” he says. “The problem there is carbon reduction. The data center is a huge power consumer. “
According to Kishimoto, data center industry groups in Japan value PUE but there is a cultural aversion to the amount of options in the ways it can be measured. “Japan is totally sold, however they cannot stand the very informality of the definition,” he says. “Japan is a very precise country. When they see something becoming a standard they’ll beat it to death. They study it and try to define it in a precise fashion.”
PUE - SUCCESS IN INFORMALITY
In Kishimoto’s opinion, informality of the metric’s definition was partially responsible for its success. “PUE was very informally defined,” he says. “Because of that, it was very easy to get a grasp of what’s behind it. PUE is a compromise.” Monroe agrees: “PUE is popular because it’s understandable and it’s relatively cheap to get a number.”
Ultimately, Kishimoto thinks it would be beneficial to have a single international metric for data center energy efficiency and PUE has already gone a long way in that regard. “It’s very nice to be able to have something like that,” he says. “If you’re talking about PUE in a very informal sense, it’s already a very international metric.”
A potential obstacle in further acceptance of PUE by the market is the effect of the type of applications a data center supports in the design of its infrastructure. Some facilities have to have more redundant and secure infrastructure than others and the level of infrastructure redundancy is often directly proportional to the facility’s overall energy efficiency.
There are multiple ongoing efforts to define a unit of useful work in the data center and measure the amount of energy it takes to perform it, but different applications do different types of work, making this a difficult task. Some define it in terms of the number of users served, some define it in terms of the
THE GREEN GRID IS WORKING ON ITS PROXIES FOR MEASURING USEFUL WORK OF THE DATA CENTER.
The organization sees the development of these as the next step in providing metrics that can be applied to the output of a data center – an attempt to marry units of work to total facility energy. A green grid technical committee was looking at the useful work.
The GG splits the proxies thus:
DceP-based Proxies Useful Work Self-Assessment and Reporting: All applications self-report, utility function is uniform and equal to one; user must decide how to normalize each task.
DCeP Subset by Productivity Link: A subset of all applications self-report using Productivity Link, a freely available Software Development Kit (SDK); scale results to represent whole data center.
DCeP Subset by Sample Workload: An instrumented subset of servers is measured running as sample workload; scale results to represent whole data center.
Bits per Kilowatt-hour Sum all outbound bitstreams from data center, divide by energy used by data center.
CPU Utilisation Based Proxies Weighted CPU Utilization – SPECint_rate Use CPU clock speed, SPECint_rate benchmarks, and CPU utilization to determine amount of work being done.
Weighted CPU Utilization – SPECpower Use CPU clock speed, SPECpower benchmarks, and CPU utilization to determine amount of work being done.
Compute Units Per Second (CUPS) Uses trend curve based on Moore’s law, age of server, and CPU utilization to determine work done.
In the end, there will probably be more than one proxy:
• Homogeneous data center: DCeP-based
• Heterogeneous data center: CPU or bit- based.
number of transactions processed and others define it in terms of the number of CPU cycles performed. In Monroe’s opinion, the most useful international metric for data center energy efficiency will apply equally to most data centers, regardless of the application, from telephony to enterprise computing.
See more articles on metrics at
www.datacenterdynamics.com
www.datacenterdynamics.com 27
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60