News Stay informed about the latest enterprise technology news and product updates.

Service providers ponder cost of data center buildout options

There are a lot of ways to skin a data center infrastructure buildout, service providers say.

Faced with building out a large data center, budding cloud and managed service providers face a stark choice: They can opt for low-cost, commodity servers or for pricier proprietary blade environments like Cisco's Unified Computing System.

For as long as there have been hosting providers, the gavel for data center buildout usually comes down in favor of low-cost vanilla systems with no proprietary parts. But IT architects say some manageability features of new proprietary systems tip the scales in their favor, despite their higher initial price tags.

Everyone knows what side Google, the archetypal cloud provider, is on: It runs low-cost, custom servers built to deliver exactly what Google wants and nothing that it doesn’t. Google is notoriously tight-lipped about the make-up of its data centers, but it’s probably safe to say it doesn’t run many Cisco or IBM servers.

The low-cost model for data center buildout
SoftLayer, for one, adheres to this basic Google philosophy. The Dallas-based Infrastructure as a Service (IaaS) provider is no Google, but it’s no mom-and-pop shop either. With Web hosting, virtual private servers and, more recently, multi-tenant, shared hosting offerings, it runs 30,000 chassis across three U.S.-based data centers and adds about 1,000 servers per month, said George Karidis, SoftLayer’s chief strategy officer. To accommodate this growth, SoftLayer partners with white-box manufacturer Super Micro, currently the fourth largest server vendor by unit after Dell, Hewlett-Packard Co. and IBM.

SoftLayer’s reasoning for this approach is threefold, Karidis said. For one thing, there’s cost. “The chips, processors, RAM and drives are all the same, but the overall chassis cost is less,” he said.

Then, from an engineering perspective, there’s flexibility. “We’ve requested specific requirements on the board and the chassis,” he said, for example more efficient power supplies, or taking out unnecessary USB ports that draw power.

Finally, there’s standardization. “We know every month exactly what we’re getting; that’s critical to our market,” Kariadis said. Going with commodity servers, the provenance of internal components is sometimes left up to the discretion of the OEM. A server might ship with a Seagate hard disk drive one month and a Hitachi drive the next. But that kind of inconsistency doesn’t work for SoftLayer. “This way, we can eliminate potential troubleshooting issues down the line,” he said.

Cloudosity in Australia is a fraction of SoftLayer’s size – it currently counts five people on its staff -- “but world domination is on our hopes and dreams list,” said David Cheal, the director and technical architect. As such, the IaaS startup thought long and hard about its server choice for a data center buildout.

They say a blade is a blade is a blade, but clearly it's not.

Chris Gabriel, director of solutions and marketing, Logicalis Group

“There are a couple of different philosophies on how to build out a data center,” Cheal said, with some architects preferring for the reliability of high-end IBM and HP boxes. “But for us, it’s such a distributed environment, that we can afford to lose a box and go with more expendable Dell kit,” he said.

However, for storage, Cloudosity opted for relatively premium NetApp Network File System storage, which was justified by the space savings offered by its primary data deduplication capabilities, he said.

Premium features, premium pricing
Other wannabe-cloud providers see the logic of investing in proprietary hardware to exploit – and resell -- unique features.

In its quest for gear to host cloud-based disaster recovery services, U.K. managed hosting provider Logicalis was impressed by Cisco’s Unified Computing System (UCS).

“UCS just blew us away,” said Chris Gabriel, Logicalis’ director of solutions and marketing. “They say a blade is a blade is a blade, but clearly it’s not.”

What got Gabriel excited was “the transient nature of the [UCS] blade that lets you do some really clever things,” he said. “It allows us to think of infrastructure in a completely different way and sell and make money as a service provider.”

For example, the stateless-UCS blade allows Logicalis to offer higher-level disaster recovery (DR) services than is otherwise available. Using CA Spectrum Automation Manager atop Cisco UCS, “We can automate and pick up the bits of [a customer] data center, move it to our data center, and have it up and running in four to six minutes,” he claimed. “We can stand up Oracle and other database applications looking exactly like they did in [the customer] data center.”

On other platforms, reliance on networking MAC addresses precludes that level of flexibility, Gabriel said.

However, this high-level functionality requires that Logicalis customers also run UCS in their own data centers, thereby limiting the number of organizations that can take advantage of the service.

The peculiarities of the U.K. economic and political situation provide an answer to that particular problem, another European systems integrator explained.

“With the economic downturn, a lot of local and national organizations are under pressure to work together to use a shared-services model,” said Tony Conway, vice president of services at MTI, a data center infrastructure consulting firm.

Specifically, Conway said MTI has a rich pipeline of business from public agencies that are considering the VCE Coalition’s Vblock – a combination of VMware virtualization, Cisco servers and networking, and EMC storage – to perform virtual desktop infrastructure, or VDI. “Platform sharing is a huge driver around here, and [Vblock] lends itself well to that use case.”

Logicalis’ Gabriel sees that same dynamic at work in sectors such as higher education and healthcare and hopes that that will translate into a market for its UCS-based disaster recovery offering.  For example, Gabriel described a scenario where four hospitals each have their own private UCS and storage, but all replicate their data back to a single set of disaster recovery infrastructure at Logicalis.

It goes beyond the idea of simple shared cloud infrastructure as a service, Gabriel said. “We talk about infrastructure zones – private community infrastructure shared by like-minded individuals.”

Let us know what you think about the story; email Alex Barrett, News Director, at, or follow @aebarrett on Twitter.

Dig Deeper on Server hardware strategy

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.