Data center executive discusses pivotal role of Cisco EnergyWise in development of a thermoelectric "backplane" for dynamic energy managementFebruary 09, 2009
February 09, 2009
Kevin Smith says running data centers requires connectivity, power, and "lots of know-how." With some help from Cisco EnergyWise, Smith is using his extensive know-how to create an energy management system that could revolutionize the industry.
Smith is the chief executive officer of Global Access Point, one of the leading providers of "carrier hotels" and data centers in the United States. Many of the country's largest telecommunications companies, including AT&T, Verizon, Qwest and Level 3, use Global Access Point's facilities to interconnect with other carriers. Dozens of businesses and academic institutions also use his operations throughout the state of Indiana and in Chicago.
Certainly, data centers could use some help in managing their energy usage. All data centers worldwide already use 0.5 percent of the world's energy. Experts predict that data center electrical usage will increase roughly fivefold by 2020 if current growth patterns and energy practices continue. The greenhouse gas emissions caused by such rapid growth is foreboding. The world's data centers combined already generate as much pollution as some countries.
"Doing the math shows some serious challenges ahead of us regarding electrical energy."
But Smith is hard at work on a way to greatly reduce the energy demands of data centers. Smith, who studied psychophysiology in college and has owned a software company, is working with Cisco and some of the region's leading research institutions, including Notre Dame University's supercomputing program, to create a new system for energy management that aims to stop the power drain in data centers. He is now filing three patents related to the technology and concepts created for this system.
News@Cisco recently spoke with Smith about his efforts, which he expects will save his company as much as 50 percent of its energy costs within a few years.
How did you first start working with Cisco and its EnergyWise technology?
Kevin Smith: For the past couple of years I've been developing my own system to better control energy usage in our data centers. I began building this system from the ground up, since it appeared there was nothing on the market - or likely to come to market - that provided this capability.
As part of this effort, I realized I would need to build something to track energy usage. Then one thing lead to another, and I discovered Cisco was developing EnergyWise. I've been working with Cisco for the past few months as a beta tester. I'm absolutely ecstatic about EnergyWise because it provides a ready-made, standardized framework of information that can feed into the energy management system I'm developing.
How will your new system work with Cisco EnergyWise?
Kevin Smith: EnergyWise will be the cornerstone for providing information about electrical usage in all my company's data centers and points-of-presence. As they say, you can't manage what you don't know, and EnergyWise will give us the ability to "see" exactly where our energy is going. By knowing one factor how end devices such as servers are using energy we can calculate the rest of the factors, such as heat gain and cooling requirements, in any area of our facilities.
EnergyWise is making it possible for the first time to map our energy usage to the finest granular level, the endpoint device. This means that we can account for almost all sources of energy use. That offers incredible visibility into the energy activities in our data centers. Information gathered by EnergyWise becomes the basis for making decision about how everything should be controlled. What I'm focused on is building an extended system of intelligence that can analyze information collected by EnergyWise and make the automated, proactive decisions necessary to direct energy to exactly where it is needed at any given moment.
EnergyWise will play another crucial role in our business. It will be relatively easy to link EnergyWise to heating, cooling, lighting and other components traditionally beyond the control of the network. It is something we will use to reduce energy consumption in our headquarters building as well as our computing facilities.
EnergyWise is really the nervous system for my energy management system. The knowledge engine I'm developing becomes the brains. EnergyWise provides the feedback on physical electrical activity in a data center. The "brain" processes that information then triggers the nervous system to respond appropriately by turning off or powering up devices only as necessary.
How will the 'brains" of your new energy management system work?
Kevin Smith: My system will take the information gathered by EnergyWise and use fuzzy logic to learn and then dynamically and proactively respond to the energy needs of each component in the data center. The concept is to look at a data center or any building, really as an energy envelope. The idea is that once energy enters the building, it needs to be used and reused as much as possible. You should only let it escape if absolutely necessary.
For data centers, electricity is usually converted to heat as it is drawn through servers and other electrical devices. But heat is also energy, just in a different form. While I can't convert that heat back into electricity, I can use it to avoid drawing more energy. So heating is tracked just as closely as electricity, and I can find where I have excess and where I need more. I can then move that heat to where it's needed, perhaps even storing reserves of thermal energy.
The beauty of EnergyWise is that it allows me to extent this energy management system as far as the network can reach. With EnergyWise, I can monitor and control energy usage among all my buildings, not just in one building. I think of my system as a thermoelectric "backplane " for my entire business.
What is the ultimate goal for your energy management efforts?
Kevin Smith: The main goal is to get the same amount of computer support with much less energy. This is also the basic concept behind the virtualization of computer resources. Rather than running 12 servers for 12 applications, why not use just two servers that share the work of running those 12 applications? That eliminates a lot of waste.
I'm approaching our energy management in a similar way. By using more sophisticated management software, I want to provide just the right amount of energy to each computing task, and no more. If a rack of servers doesn't need to be on, then they shouldn't be on. If another rack needs to work extra hard, then it should get the electricity it needs to do its job. But once it's done, the energy should be turned down accordingly.
The machines should not keep buzzing if they aren't doing any work. It's analogous to people turning off their cars. They don't leave them idling in their garages overnight. Yet this is exactly what we do with many electrical devices in data centers. My aim is to create a system that can proactively turn off any device that isn't doing work at that moment.
So the goal is to create an "easy button" that makes turning all of these devices off as simple and automated as possible. The trick is to build a system with the necessary intelligence to manage all the dynamic uses of energy by these thousands of devices. Some bad things can happen if you just pull the plug on computers. Electrical devices need to be turned off in the right order. And since there are so many interconnected and interdependent machines in data centers, it is crucial that all of this is done very carefully. This is where most of my work is focused.
How does your energy management system fit with other efforts to reduce the costs of running data centers?
Kevin Smith: It really goes hand-in-glove with virtualization efforts. Just as virtualization software allows data centers to better utilize computing resources by allowing administrators to see where they have extra capacity and then gives them the means to very carefully manage that capacity, EnergyWise coupled with a system like mine can do much the same for energy. It's the idea of using resources only when and where you need them. I think until recently everyone has looked at energy and computing resources as fixed costs and static elements. But, in fact, they are highly controllable and highly dynamic factors in data centers. We just need the right tools to management them properly.
Why is better energy management so important to your business?
Kevin Smith: Well, doing the math shows some serious challenges ahead of us regarding electrical energy. The energy required to run a data center is extremely concentrated. For example, my main data center in Union Station, Ind., requires about 10 megawatts of power. Downtown South Bend, Ind., draws about 13 megawatts. And just one typical server rack has similar electrical demands as an entire home! So keeping this kind of intense energy usage under control will be crucial for all concerned. It's a big expense and these days we really can't afford to waste any energy. And all of this is tied to greenhouse gases, which of course is another major concern for the entire world.
Also, energy usage in data centers is highly dynamic, and, as is well known, a lot of data centers are designed to accommodate rare but intense peak usage of Internet sites. And even with non-commercial computing systems, such as Notre Dame's supercomputer, electrical usage can fluctuate by 40 percent.
Until now, electrical provisioning for data centers has been based on static, fixed systems. Such a static system is always running a step behind dynamic use. With EnergyWise and a sophisticated, holistic management system, you can anticipate usage patterns and automatically provide just the right amount of electricity when and where it is needed.
My long-term goal is to be able to efficiently support grid computing and massive parallel computing. To do that, you need a lot of resources working in unison. Comprehensive energy management will be an important part to grid computing because that is where the network and computers really merge into one.