Anatomy of a Cisco data center convert
- 31 July, 2010 04:49
Cisco almost didn't make the cut. NightHawk Radiology Services was ready to hand its data center consolidation project to another vendor until a facility leasing issue delayed its decision.
That gave Cisco an opportunity to pitch its Unified Computing System (UCS) proposal to the lab, which provides radiology services to 1,600 healthcare sites – 26% of all hospitals in the United States. NightHawk conducts 3 million studies per year with an average turnaround time of less than 20 minutes per study.
"Our project did not include UCS at the outset," said Ken Brande, vice president, information technology at NightHawk. "A facility lease issue delayed the project and in that delay period Cisco presented UCS."
Enter Cisco. NightHawk will replace 120 physical IBM and Dell server blades with 18 UCS servers in four chassis running virtualized workloads. Granted, it's not a Google or Amazon.com or Yahoo scale data center with thousands of servers – but NightHawk fits the prototypical profile of a UCS target: one looking to consolidate, virtualize and reduce complex operations.
The NightHawk project entails consolidating ad hoc server rooms in multiple sites globally, reducing cycle times of new technology deployments, reducing capital and operational expenditures and simplifying management – all in support of handling 9,000 studies and 500Gbytes of image data per day, and 120 Terabytes of online data.
"The environment was difficult to manage," Brande says. "When you have a bunch of disparate technology resulting from 'Get the best for what we can get for the price today, and then push it out as fast as possible,' it means that a lot of these spaces didn't have the appropriate power."
NightHawk IT was focused more on maintaining that type of environment rather than being responsive to new business requirements.
"It was hard to build a new business product with technology that was dispersed for no rhyme or reason," Brande says. "And then the resources were too busy maintaining the assets and not available to help drive the new business."
So NightHawk, working with Cisco systems integration partner World Wide Technology, embarked on its server room consolidation/data center strategy project last year. The lab sought to put its data center resources on a common platform with consistent management tools that let it rapidly deploy new technologies and services.
"We wanted to minimize the number of issues that happen, be able to recover from them faster, and just have basic administration of your environment be a task that is easy to do and doesn't consume you," says Christopher Smith, manager of data center infrastructure at NightHawk.
NightHawk was attracted to UCS Manager, which the officials said presents a single intuitive interface among applications from Cisco, BMC Software and other partners. Alternatives offered less coherency among applications – the seams between applications from various independent software vendors were evident even though they may have been accessible from a single interface, the NightHawk officials say.
"We had hundreds of very disparate systems from different vendors with quite different ages and different classes of machines," Smith says. "There's quite a lot of difference in the hardware that made it a challenge to manage."
Not that UCS was a total cakewalk either. NightHawk IT had to learn the concept of service profiles in which server deployment and workload allocation, configured through templates, is divorced from the physical servers themselves. Servers are "built" from these templates and the profiles can exist on any blade at any time.
Service profiles are intended to define and enforce resource access privileges for virtual machines, and follow VMs as they move across and between data centers.
"The idea of separating out the configuration of the hardware from the actual hardware itself" required some instruction, Smith says. The rest was just months of burn-in testing of myriad data center scenarios, online documentation review, policy definition and practice, practice, practice.
"Once you've got that in line the rest is pretty straightforward from there," Smith says.
And the results are evident. NightHawk has cut its cabling and switch port requirements to a fraction of what they were: 24 connections – 16 network and eight Fibre Channel -- were reduced to four 10Gbps Fibre Channel over Ethernet and four Fibre Channel with UCS, Smith says. The installation requires only 20% to 25% as many upstream switch ports to support the same level of performance and connectivity, he says.
Server expansion cost is 25% to 33% that of alternatives, NightHawk claims. Server deployments and updates have been reduced to minutes from hours.
NightHawk expects payback from the investment in about three years, including a 12% annual reduction in operating expenses.
Read more about data center in Network World's Data Center section.