By Brad Thompson
There are numerous new challenges influencing data center design today, caused by a multitude of changes in the way companies do business. New trends such as server consolidation and advanced technologies such as blade servers and VoIP have created increased power and cooling requirements in the data center. With the ever-increasing global, 24-hr business economy, it means that even the smallest amount of downtime is not an option. And in the post-911, post Katrina world, companies have a heightened focus on threats to their business continuity. All of this, combined with the shriveled IT budget of the post-tech bubble world, has created quite a paradox in the approach to data center design. There is now a bigger need for more power, more cooling and higher availability, and a smaller amount of budgeted money to provide it.
One of the big changes in the IT environment that affects data center design is the trend of server consolidation, which is the reduction of “server-sprawl”, a situation where there are too many under-utilized servers consuming more resources than necessary. Over the years, the low cost of computer server equipment combined with the practice of dedicating servers to single applications has led to widespread situations of server sprawl. It is estimated that servers in many companies are running at 15-20% of their capacity, leaving a majority of that server’s capacity untapped, but yet still taking up space and consuming power. The solution to this is server consolidation, or utilizing fewer servers to perform more tasks. A new technology that is being used to facilitate this trend is the high-density, blade server.
A blade server is a server chassis that houses multiple ‘blades’, or circuit boards that each act as an individual server. This allows more processing power in less rack space, simplifying cabling and reducing power consumption. There are many obvious advantages of deploying blade server technology, including resource consolidation, load balancing and failover capabilities, but this all comes at a less than obvious price, increased heat output. Since blade servers pack more processing power in a very small footprint, the heat generated is much greater and highly concentrated, thus requiring special cooling infrastructure. The biggest problem in blade server deployment is a lack of proper cooling. A blade server which is not properly cooled will re-breathe its own hot air and overheat. Most legacy data centers have raw power and cooling capacity but aren’t equipped to deliver this capacity to a high density area. Fortunately, there are new cooling systems available such as the InfraStruXure RC by APC, which is designed to sit directly next to the IT equipment rack and can deliver up to 60kW of cooling per rack. This cooling approach design is referred to as in-row, which turns the rows of IT equipment facing away from each other, so the backs of the equipment direct their heat into one row which is then cooled by a unit like the one mentioned above. This approach prevents the hot and cold air from mixing, which increases the efficiency of the cooling unit and increases its cooling capacity. Yes, you can save real estate by deploying high density servers, but you must be prepared to protect your investment by employing high density cooling.
In the past, when building a data center, the trend was almost always to install a costly raised floor system, which provided cold air distribution channels and space to run power and data cables underneath the IT equipment. The raised floor system was adequate for providing the cooling and cable management for data centers with a small number of large IT devices that didn’t change very often. But today’s data centers have a higher number of small IT devices, so the number of branch circuits per square foot is much greater, and devices are being changed around regularly. This increase in cabling now creates a congested cooling channel that is inefficient as well as cabling that is more difficult to access, thus resulting in delays and costs associated with changes in the data center. Also, with more devices occupying less space, the lbs per square foot is much higher, creating a weight capacity problem for raised flooring. With some racks potentially reaching weights of 2,000lbs, special reinforcing may be required in a raised floor environment, requiring another significant increase in cost and planning. These are just a couple of reasons why the raised floor configuration, which went unchanged for almost 40 years, is now starting to be viewed as impractical and inefficient for today’s high-density and high-availability data centers. The money that the companies can save from expensive raise floor configurations will be better used in other areas of the data center, such as in-row cooling architecture, overhead cable management, redundant UPS’s, remote management and security.
One of the biggest mistakes made in designing a data center is oversizing, which is the act of installing more power and cooling capacity than can be utilized, thus creating a much larger TCO (Total Cost of Ownership) than necessary. Instead of oversizing by buying extra capacity and expecting to grow into it, it is best to design a data center using a standardized, modular, building block approach, that can adapt to growth and changing technologies.
Brad Thompson is with Bradford Technology Group, a provider of consulting, design & installation of network critical physical infrastructure. www.bradford-tech.com
© 2006, TechJournal South. All rights reserved.
