|
 |
|
General Information
|
Add To My Personal Library |
July 2, 2010
• Vol.32 Issue 14
Page(s) 19 in print issue
|
Data Center Metrics Demystified
If You Don’t Measure, How Can You Manage?
|
| Key Points
• Some of the confusion surrounding data center metrics such as PUE (power usage effectiveness) and DCiE (data center infrastructure efficiency) stems from the fact that they are advertised as “energy efficiency metrics” for the entire data center, which is not the case. Both metrics can determine only how much energy is lost in non-IT facilities.
• Without metrics, data center managers are blind to where energy is being consumed and the impact of any actions they may take.
• DCeP (Data Center energy Productivity), which measures the ratio of useful work produced to total data center energy consumed, is the latest metric proposed by The Green Grid. | Data center efficiency goes beyond knowing if servers are still up. These days, data center managers are accountable for energy usage, energy efficiency, compliance, regulation, and the list goes on. Performance must be monitored and trends must be predicted to ensure that the data center is always up and ready for capacity increases at any time. For this to happen, a well-designed set of metrics must be enforced. But there’s plenty of confusion surrounding data center metrics, especially metrics such as PUE (power usage effectiveness) and DCiE (data center infrastructure efficiency).
Here, we break down some of the leading data center metrics in use in small to midsized enterprises, what they mean, and how knowing more about them can impact your enterprise.
 Why The Confusion?
Daniel Kharitonov, senior staff engineer at Juniper Networks (www.juniper.net), says the confusion primarily stems from the fact that PUE as well as its reciprocal, DCiE, are advertised as “energy efficiency metrics” for the entire data center, but he says this is simply not the case. Kharitonov explains, “Both can determine only how much energy is lost in non-IT facilities, such as power transformers, the electric grid, and cooling systems. By definition, ‘efficiency’ is the ability to perform useful work at minimum cost point.” He says if the data center cannot handle transactions at a minimum energy footprint, it does not matter how good the facilities are, because energy is still wasted with a PUE of 1.
Dr. Joe Polastre, CTO and co-founder at Sentilla (www.sentilla.com), says confusion exists because PUE and DCiE are not a measure of the data center’s efficiency; rather, they are a ratio of total power consumption to IT power consumption. Polastre notes, “A standard physics textbook will tell you that ‘efficiency’ is defined as the amount of work completed for each unit of energy consumed. PUE and DCiE do not measure efficiency.”
The Metric Lineup
Larry Vertal, executive director of The Green Grid (www.thegreengrid.org), says that small and midsized enterprises today are the ones least likely to be using metrics. He says this is because staffing issues are acute in SMEs. He says many of the metrics used today by SMEs aren’t even taking power and energy into account at all. Vertal comments, “Many are homegrown tools to measure CPU utilization, SPEC-ratings, [and] runtime duration measurements and don’t take into account energy usage. Effectively, they only consider half of the issue of energy-based productivity.”
Vertal says in some cases, data center managers try to guess at power usage and efficiency based on equipment nameplate ratings, which, he says, are highly divergent from the actual power or energy used.
Tate Cantrell, CTO of Verne Global (www.verneglobal.com), points out that beyond PUE and DCiE, there are other opportunities to quantify success. Cantrell says, “Uptime, for example, can be calculated as the percentage of time that the power is on or can even be analyzed as the percentage of time that stakeholder applications are online. Bandwidth and network utilization are also important metrics that show trends for how information is used within a data center and how a data center communicates with its stakeholders. And power utilization is an important metric that shows a data center manager what percentage of the facility is available for additional capacity.”
Cantrell also says that average server utilization is an excellent way to show a data center manager how effectively the capital resources are being utilized within the data center. “Just because a low-PUE data center is effectively cooling the watts that are used to create information does not mean that the computational watts are being utilized effectively,” Cantrell notes. “Through technologies such as virtualization, a data center manager can drive up average server utilization and improve the DCeP (Data Center energy Productivity).” Cantrell says that DCeP, which measures the ratio of useful work produced to total data center energy consumed, is the latest metric proposed by The Green Grid.
According to Kharitonov, an ultimate data center energy efficiency metric is how much it costs (on average) to run a transaction. He adds, “However, data center designers often do not have the end-to-end visibility early into a project and therefore need to consider DC components individually. At each component level, there are metrics for energy efficiency assessments. I would single out SPECpower (for servers), ECR (for network equipment), and PUE (for non-IT facilities). Together, these three metrics (or their functional equivalents) define the majority of energy uses within a data center.”
W. Pitt Turner, executive director of the Uptime Institute (www.uptimeinstitute.org), says having competing metrics discourages the industry from taking any action. He notes, “The Institute, DoE, EPA, The Green Grid, and other organizations in the data center industry agreed in January 2010 to focus exclusively on the PUE. With the singular focus on PUE, this distraction should be eliminated.”
Knowledge Is Power
So how can knowing more about these metrics impact the enterprise? Turner says without metrics, operators and managers are blind to where energy is being consumed and the impact of any actions they may take. He elaborates, “Using metrics will help identify areas of opportunity and indicate the progress made. In many IT organizations, the CIO is not responsible for paying the cost for energy. Yet, the decisions made by the CIO have a direct and profound impact on the utility bill. The CIO needs to be accountable for the cost of energy at a data center.”
In Polastre’s opinion, the data center is transforming to an industrial manufacturing facility, where the product boils down to compute cycles. He further explains, “If you view the data center in this way, optimizing data centers is a matter of process efficiency—how do you minimize the amount of energy in [and] waste and heat out and maximize the amount of work done? Benchmarking and optimizing applications for metrics like MIPS/watt and transactions/kWh lead to the overall system being more efficient, both facility and IT system.”
Vertal concludes, “In a nutshell, if you don’t measure it, how can you really manage it? Unless there is a sudden breakthrough in practical nuclear fusion, you should bet your business on the fact that energy costs will continue to increase. And as a result, the operational costs of data centers are becoming greater than the IT procurement costs. While the approaches may not be uniform by governments globally, we have already seen regulation emerging that includes data center footprints. Measurement and metrics become essential to address the emerging accounting systems.” 
by Chris A. MacKinnon
PUE Explained
According to Tate Cantrell, CTO of Verne Global (www.verneglobal.com), there is minimal confusion surrounding PUE in the industry. He says The Green Grid introduced the concept in 2007 to wide acceptance and has continued to educate the industry about the importance of data center metrics in continuous improvement of data center efficiencies. He notes, “The PUE is very simply the ratio of the total facility power to the IT equipment power (servers, storage, and network). Regardless of what measurement point is used in the PUE calculation, the fact that the majority of data center managers are now recording this metric on a regular basis and working to improve upon it meets the ultimate objective of the parameter—improvement.” |
|
|
