Image courtesy of Shutterstock.
When teams and tools work across silos, the synergy created becomes the basis for competitive advantage. Gathering good data streams — metrics that matter to both business and IT — and correlating them through powerful analytics will amplify bottom line results. As an example, by measuring and analyzing more than just power utilization effectiveness (PUE), the focus of continuous optimization shifts to risk reduction, revenue growth, decreased capital and operating expenditures, and enhanced customer experience.
Here are 5 key ways more efficient power utilization can enable data centers to be more effective and efficient.
Don’t Make Data Centers become Highly Efficient Waste Centers.
The opportunity and imperative to optimize energy use in the data center industry is at an all-time high. In the modern digital era, data centers are as essential as power plants, a massive and critical infrastructure upon which our social, business, retail, healthcare, and government services are run. In the US, data centers are the fastest-growing consumers of electricity. American businesses spend $13 billion annually to power data centers, and the costs don’t stop there — the 150 million metric tons of carbon pollution emitted annually by data centers is a significant contributor to pollution and climate change.
A report recently released by the Natural Resources Defense Council (NRDC) points out that while hyperscale cloud data centers like those run by Apple, Google and Facebook get a lot of attention, they have been primarily optimized for energy efficiency and account for less than 5% of US data center electricity consumption. The other 95% is consumed — and wasted to an alarming degree — by small, medium, corporate, and government data centers. One key source of waste the NRDC and others have identified is the estimated 12 million servers in the US that are doing little to no work but still drawing power. By NRDC estimates, the average server operates at 12-18% of capacity; they have argued for the adoption of a simple server utilization metric across the industry. They also advocate for the need to align incentives (i.e., close collaboration between those paying energy bills and those making efficiency decisions) and to require public disclosure of data center efficiency and carbon performance. Takeaway question one: Are resources really needed — when and why?
Getting Beyond PUE.
One of the most common measures of data center efficiency is Power Utilization Effectiveness (PUE). Simply put, PUE is the ratio of the total amount of energy coming into a data center to the energy that reaches and is used by IT equipment. Energy used by computing equipment is considered productive, while energy used for cooling, lighting, and other auxiliary purposes is considered waste. Introduced by global consortium Green Grid in 2007, PUE has served its purpose in establishing a common metric and language for energy efficiency initiatives. However, it is a relatively one-dimensional metric that has not kept pace with the evolution of data center technology. The industry requires additional useful information to continue driving optimization efforts. The major drawback of PUE is that it doesn’t address data center productivity; knowing how efficiently you are powering IT equipment tells you nothing about how much productive work that equipment is doing in service to business objectives. Takeaway question two: How much work is getting done?
Auxiliary Variables: The Impact on PUE of Cooling, Layout, and More.
The auxiliary, or waste, side of the PUE ratio represents a land of opportunity for data center managers who would rather spend their resources building innovation, growth, and agility into their operations than squandering money and electricity on idle servers and inefficient cooling. The virtualization of data centers makes it possible to gather and analyze more granular and real-time metrics from every component of the data center and pinpoint areas of inefficiency.
Advanced analytics can use PUE metrics in correlation with other business and IT metrics to gain objective insights into areas of the data center where there’s room for improvement. Even relatively isolated tweaks to airflow, cooling systems, floor layout and CPU utilization can have major impacts on efficiency and productivity when combined to achieve intelligent synergies. Let’s take just one auxiliary component — cooling efficiency. Most simply, if CPU utilization is inefficient, more power than necessary is being used, and therefore more heat than necessary is being generated. The power being used for cooling systems could be reduced if servers were turned off, or not “wasted” if the server capacity was properly loaded. But cooling efficiency also involves much more complicated issues around airflow that are difficult to measure and manage. In the context of data centers, air is a fluid, acting like a liquid as it takes the path of least resistance. In other words, cool air pushed towards hot equipment will move around the equipment in an inefficient fashion. Various approaches to channeling the cool air (e.g., hot row/cold row, plenums, etc.) can help, but are complex. By nature, airflow behaves in a non-linear fashion. Computational Fluid Dynamics is now being applied to understand and predict these types of efficiency — another example of using advanced analytics and data science to amplify bottom-line results.
The layout of the data center floor and rackspace, also closely related to cooling efficiency, is another component on the auxiliary/waste side of the PUE ratio that must be taken into account. It is inefficient to mix higher heat and lower heat generators, or to disregard airflow when aligning equipment. Careful research into floor and rack layout must be a part of every effort to optimize a data center; again this is an area where seemingly minor tweaks can have a dramatic effect on improving PUE. Tracking these physical changes on a timelines against detailed PUE-related metrics can provide vital insight and highlight unexpected successes that can then be replicated for even greater impact. Takeaway question three: how do power consumption and cooling relate, where, when and why?
Balancing Efficiency, Performance, and Cost.
Going beyond PUE — and go beyond we must — there are many efficiency metrics that matter in developing a holistic approach to data center productivity. For example, how many cubic feet per minute (CFM) of airflow are required to maintain a given temperature? The less, the better, obviously; and it depends on technical architectures among other factors.. Again, layout and other factors will impact airflow needs. It’s important to constantly reevaluate; best practices around these metrics are constantly evolving. ASHRAE, for example, has recently raised the climate threshold from 68 degrees Fahrenheit to 80.6 degrees. Also, it is essential to track the number of business transactions per minute being processed by the total data center technology stack. Relating all such metrics to the financial costs (CapEx and OpEx) of the total data center investment gets us closer to a nuanced and integrated view of data center performance.
At the end of the day, real efficiency is getting the most work done, the fastest, at the lowest cost. The “work” is business transactions, communications, etc. The infrastructure of the data center the engine driving all this digital work, and must be optimized for performance (smarter, faster, more powerful) but in the real world, performance must be balanced against cost—financial and environmental. The balancing act is like solving multiple variable setsof equations (remember college advanced mathematics?). Lots of variables creates complexity and potentially misunderstood relationships and implications.
The more “knowns” you have in terms of variables, the easier it gets to solve the problem (equations). PUE is only one of the many variables in the productivity equation. If you don’t have enough “knowns,” you can only guess. To build up your arsenal of “knowns,” you must measure everything that matters. At a minimum, you should start by measuring: power consumption by physical servers; performance of all those servers and VMs in terms of not just utilization, but transactional work accomplished; and the CapEx and OpEx costs associated with each. Secondly, ensure that everything you measure today can be accessible to an analytics solution as you progress along the spectrum from descriptive to correlative to predictive and ultimately toprescriptive approaches. Finally, develop a plan to continuously improve towards measuring more and more meaningful variables. Repeat until super-optimized, and enjoy being in the company of innovators at the top of the ultra-efficient list. Takeaway question four: do you know all your “measurements that matter”?
True Efficiency Essential to Business Success and Sustainability.
Compounded across large facilities, and aggregated across the industry, the potential energy savings alone are astonishing. The NRDC claims that if just half of the technologically feasible savings were achieved, US data centers could cut electric use by 40% and save the nation’s businesses $3.8 billion annually. Federal legislation is beginning to target energy efficiency efforts in the data center industry, starting with calls for federal agencies to implement improvements in their data centers. As social and governmental pressure continues to mount around climate change concerns, it would behoove the data center industry to stay ahead of the curve; expansion, carbon pollution, and energy use will be more intensely scrutinized in the near future. And energy savings aren’t the only prize.
Starting with well-defined metrics like PUE that everyone understands IT and business teams can bridge traditional silos and collaborate intelligently, with game-changing results. Correlating meaningful data streams from both IT and business can form the basis for new synergies leading to competitive advantage. With fine-tuned, flexible efficiency analytics and initiatives in place, data center operators are more agile, ready to ramp up to meet spikes in demand, keep pace with growth and market demands, and respond quickly to market trends and new technologies (e.g., rapid adoption of cloud computing, digitization of “everything”). Continuous optimization initiatives shift focus from one-dimensional measurements of efficiency to a holistic view of productivity, leading to risk reduction, revenue growth, decreased cap-ex and op-ex, and better customer experience. Takeaway question five: do you have an understanding of how efficiently your IT infrastructure is performing in support of business?
About the author:
TeamQuest Director of Market Development Dave Wagner has more than 30 years of experience in the performance and capacity management space. He currently leads the cross-company Innovation Team and serves as worldwide business and technology thought leader to the market, analysts and media.