Data center size and shape for enterprise IT shops may be changing, thanks to a confluence of numerous factors. Virtualization and the rise of Internet applications that tap into cloud-based services have given rise to a number of computing services to keep those applications humming along.
Given all these changes, there is no single server configuration or network architecture that will suit every business. What works best for your enterprise can be difficult to determine given the number of choices in server and network configurations.
Keeping tradition alive
Hyperscale data centers
Part 2: Data center size, infrastructure determined by type of enterprise
For smaller data centers or more modest needs, traditional architecture still makes the most sense. It is familiar and the least costly to purchase. "But you really need to understand your applications' consumption model and requirements," said Ken Owens, cloud CTO for Savvis, a large international cloud services provider that operates multiple data centers. "If you are in a more traditional environment and have typical business applications that require limited network services, you probably want to stick with traditional Intel services, maybe adding a blade server to simplify your provisioning."
It is easy to find the traditionalists. Take EasyStreet, a Portland, Ore.-based data center provider. It uses traditional server architecture completely, buying standard Dell and HP servers to host mostly small local businesses. "We don't have any blades now and don't anticipate buying any in the future either," said Rich Bader, EasyStreet's CEO. "Even our customers that are running 50 or 100 virtual machines still don't really warrant using blades. Blades carry a cost premium that doesn't make sense for us, and the economics just don't work."
Tony Stirk is the president of Iron Horse, which operates another data center for smaller customers in Springfield, Va. He has stuck with traditional architecture as well. "Blades have a large number of proprietary parts which limit competition and make your business more dependent on a specific manufacturer," Stirk said, "You have limited configuration, networking and storage choices."
Starting new traditions
But as your data center grows, traditional architectures may increase operating costs, either for power efficiency or for supporting higher computing densities. Take Contegix, a St. Louis-based data center provider. "We are deploying converged infrastructures for both our private and public clouds," said Matt Porter, Contegix CEO. "Both of these scenarios are aligned to the slicing of large pools of computing resources for small, specific needs." He plans to continue using both traditional and converged infrastructures as appropriate for customers.
This increase in scale is one reason that software-defined networks are becoming more essential, because existing virtual local area networks can't grow quickly. The same could be said about storage area networks, too.
For IT professionals making the transition from traditional to other architectures, it could be just a matter of timing which they choose. Newer architectures are still considered as replacements for traditional ones. Gartner's Butler agrees. "We are still at a relatively early stage for multi-node [a.k.a. hyperscale] servers. Their growth has been meteoric and they are already outselling blades in terms of installed numbers, but not in terms of actual dollars yet. There is a huge appetite for these kinds of servers, and they are being included in many mainstream data centers."
And with its emphasis on the cloud and with the consumerization of IT, Savvis has changed the gear it is buying. "We have had to redesign our infrastructure to be configured on demand and be more flexible," Owens said. "Customers no longer know what they need or how they want the infrastructure to support their apps, so at Savvis we are mixing blades and SANs with multi-node servers to better meet our customers' needs and provide more intelligence at the app level to make our data centers more flexible." That said, Owens predicts that multi-node racks could account for 80% of Savvis' floor space in the near future, which is a big change from roughly that percentage of traditional and virtualized servers today.
This mixture of approaches is probably what the future portends, said Philip Molter, the CTO of Data Foundry, a major data center service provider in Austin, Texas. "If Google or Facebook had to answer to an angry customer every time an instance went down because of a 0.05% increase in failure rate, they might choose a different deployment strategy."
Let us know what you think. Write to us at firstname.lastname@example.org
About the Author:
David Strom is an expert on network and Internet technologies and has written and spoken on topics such as VoiP, network management, wireless and Web services for more than 25 years. He has had several editorial management positions for both print and online properties in the enthusiast, gaming, IT, network, channel and electronics industries.