While there is much discussion and theorizing about data center efficiency, some individuals in the industry are focused on driving efficiency on a daily basis. We reached out to Clemens Pfeiffer, CTO of Power Assure and a 25-year veteran of the software industry, for a Q&A about his thoughts on achieving data center efficiency.
Data Center Knowledge: Traditionally, data center developers designed for reliability, sometimes at the cost of efficiency. Is that still the case?
Clemens Pfeiffer: Yes, this is still the case - reliability is the number one priority. If applications fail, people lose jobs. Power savings are not a strong enough argument for increased risk and the perception today, right or wrong, is that efficiency improvements decrease reliability. Unless we can demonstrate that reliability AND efficiency improvements can go hand in hand - which they do when done properly - this won’t change. Fortunately some data centers understand this and are starting to change, but it’s certainly not common.
Data Center Knowledge: When a data center operator focuses on efficiency, where are the largest potential cost savings? What are the top costs in running a data center?
Clemens Pfeiffer: This depends. In traditional data centers, there are multiple steps for savings. To give you an idea, a PUE of 1.8 (average today) means that about 40% of the power goes to cooling. So reducing the PUE or making cooling more efficient can give you 20 percent savings, easily. After you are done you have more power available for IT, which everyone uses for more capacity. Now the next biggest savings opportunity is on more efficient servers, power management of standby systems, increasing utilization of running servers and turning off spare capacity only used during peak demand. This gives you another 40 to 60 percent savings. So it’s a question of the level of efficiency data centers have implemented that determines a path forward. We have a multi-step improvement process that starts with an assessment of the current situation, looks for improvement opportunities in cooling, power distribution, IT equipment, software, operating procedures and energy contracts, all of which can lead to huge savings. But it requires a detailed study of the current situation to produce a detailed ROI-based recommendation. Without it, it’s too generic to be all that useful.
Data Center Knowledge: What are the most effective ways to reduce energy usage and achieve efficiencies?
Clemens Pfeiffer: Reduce PUE (meaning cooling overhead), buying more efficient equipment, virtualization, automated failover procedures and including on/off capabilities in the process are the ones with the best ROI. There are more, but as I said earlier, this requires a study of the existing setup to determine what’s possible, what has a great ROI and define a roadmap of what to do.
Data Center Knowledge: One of Power Assure’s recommendations for improving efficiency is eliminating transfer switches. What are the pros and cons of this strategy?
Clemens Pfeiffer: This is not an efficiency issue by itself but an enabler for efficiency improvements. The ATS limits your electrical infrastructure into an either/or model between utility and generator power. This means the IT load can run on either power source but you don’t have co-generation capabilities and you can’t use your batteries for partial utility power subsidies. By eliminating the ATS and using a distribution bus concept, generator capacity from maintenance cycles can seamlessly integrate into the current feeds. During times of high power cost, demand response events etc, some of the battery or generation capacity can be used to offset utility power further saving money (or making money by selling it to the grid) which increases the flexibility and reduces the risk.
Data Center Knowledge: Another tip is to match server capacity to load in real time. This is a challenging issue for most data center operators. What’s the most effective approach?
Clemens Pfeiffer: It is challenging, particularly when managing it across two data centers; however, it is something VMware, IBM PowerVM load balancers are supporting. We will see more and more functionality to support this ability as matching server capacity to load in real time increases application reliability – and reliability remains the first priority. When done correctly (using automated runbooks), you can add power management as part of the automated procedures and save a lot of money, which can actually pay for the implementation.
Data Center Knowledge: What are your best tips for reducing cooling costs?
Clemens Pfeiffer: I know you don’t like to hear the "it depends" answer but a hot/cold aisle configuration and using an outside air economizer are the most common ones IF the data center location is in an area where these work and the characteristic of the data center allows for it. If you have mainframes with water cooling, racks with 50kW of power as demand, some of it won’t work due to the density.
Data Center Knowledge: What are things that can be done through virtualizing servers that can increase utilization and decrease cost?
Clemens Pfeiffer: The best and most efficient set-up is based on research and patents from PARC (The Palo Alto Research Center) around shared reservation models and QoS (Quality of Service) levels of applications. Unlike static reservations, which do not respond to utilization and therefore are based on an allocation for peak demand, grouping critical and non-critical applications on the same cluster using a shared model will provide more flexibility and reliability by allowing critical applications to take over whatever they need from the cluster and non-critical applications to take the rest resulting in a much better consolidation than the typical 14:1 reduction.
Would you like to participate in an email Q & A session with Data Center Knowledge editors? Email [email protected] and you may be selected for a expert Q & A.