The increasing demand for IT services has resulted in more and more data centers being built worldwide, providing either in-house IT services for a single organization or cloud-based services for a number of customers. The U.S. government alone has doubled its portfolio since 2011, yet along with this growth comes an increasing demand for data centers to be more energy efficient. How businesses go about measuring this efficiency is an entirely different matter; too often they lack a full understanding of exactly how well their data centers currently use energy, let alone how any changes would affect it. Techniques such as predicting energy use and managing IT equipment more efficiently can have huge cost benefits, yet organizations are still falling at the first hurdle.
Predicting and Measuring
A recent SAP report found that a lack of energy insight prevents organizations from answering critical questions about their data center operations. For example, what components consume the most energy? How does consumption change on a daily, weekly, monthly or yearly basis owing to climate variations or varying IT loads? Businesses are increasingly metering and measuring their energy use to answer these questions. Even assuming they use the metering technology available to them, however, they must also ensure that they are predicting data center performance, rather than simply measuring—a distinction many still struggle to make. Measurement and metering allows businesses to see how their data centers are performing, yet only prediction allows them to determine whether what they are measuring is right and the system is actually working as planned. Once they understand this difference, businesses can measure data and predict trends, managing costs, planning investment and uncovering any data center inefficiencies. If organizations can both monitor and predict their data centers’ energy use, they can begin to address their primary concern—ensuring that the data center is operating as it should be.
The Big Questions
Once a business has made sure it can predict energy use or, in other words, determine the energy it should use, as well as measure it, there is still the question of how to decipher the information available. Correctly monitored data centers now produce huge amounts of information; the challenge for businesses is knowing what information is relevant to their understanding of energy use. Many become paralyzed at this point. Either they try to measure everything and find themselves overwhelmed, or they ignore data that would provide vital insights—meaning that any prediction or measurement will merely be a waste of resources.
Many data centers are designed with vast amounts of metering points, many of which are unnecessary. Businesses must strike the right balance to avoid any wasted efforts. For example, metering relies on the principle that businesses will be able to gain vital information on the performance of their data centers. Therefore, it is essential that organizations first work out exactly what the crucial factors are in energy performance, ensure they can predict what those factors should be and then put metering in place to be able to monitor and validate those predictions.
Understanding the Environment
To identify these crucial factors, businesses must understand the environment that creates them. Understanding the data center environment can help them avoid spending resources on metering when it’s unneeded. There is the example of a team at GE that placed thousands of sensors to troubleshoot a manufacturing process before discovering that just one, single, key sensor identified when the process was in the optimum state. Having a clear picture of exactly how a data center fits together and which points of measurement are of most significance for predicting and monitoring energy use is a crucial first step. That is in addition to the significant savings made by not installing and not having to manage unnecessary meters.
Such knowledge also allows organizations to recognize exactly what a meter is telling them. For example, if a meter goes up or down over time, was that change expected? And at that scale? Again, the key is understanding the system and its sensitivities rather than gathering data for the sake of it. Having a model of the system and being able to predict how it should behave in given circumstances is essential. For instance, if an organization understands its software workloads, it should be able to identify precisely whether a meter’s change in reading is due to the predicted software load or another factor.
The ability to predict and then measure data center efficiency and performance can have benefits beyond reducing energy costs; for example, it can be easy to spot when a component is using far more or less energy than expected, and so may be at fault. Cost, however, should still be the primary focus. In a rapidly changing data center market, metering and measurement tools must now be recognized as valuable tools for businesses; yet at the same time they are only part of the solution. To gain complete control of their data centers, businesses must be smart about what data to collect and how to process it.
Success rests on four simple steps: decide what information is needed, accurately predict what that information should be, identify what to measure to get this information and then forecast and monitor to ensure expectations are met. Anything else is simply window dressing.
About the Author
Zahl Limbuwala is CEO and cofounder of Romonet and is passionate about the data center and ICT industry. Starting his career as a chartered electrical engineer, he then moved into IT systems engineering, network engineering and later software development. Zahl is a chartered engineer, chartered IT professional and Fellow of the BCS. He has 20 years of experience in companies such as Digital Island, Real Networks, Cable & Wireless and many engineering firms writing control and automation software for production-line automation and robotics.