Cisco Dives Deeper Into Blade Server Strategy

blade server data center virtualization processing

On March 16, the San Jose, Calif., networking giant unveiled UCS, a data center infrastructure architecture that unites compute, storage access and virtualization resources into one system. UCS ties together Ethernet switches, blade servers and a one-stop management console and features a flurry of contributions from technology partners. At the time of the launch, Cisco CEO John Chambers said UCS will "bring together the compute power, the storage access and the networking capabilities" of the next-generation data center and facilitate Cisco's vision of accessing any information from any device, any network and any location at any time.

UCS, which is expected to ship by the end of the second quarter, comprises the UCS Manager, embedded software that manages the system; the UCS Fabric Interconnect, appliances that offer either 20-port or 40-port 10 Gb Fibre Channel over Ethernet; the UCS Fabric Extender remote line card; the UCS Blade Server Chassis; and UCS Virtual Adapters. The system, Cisco has said, offers embedded management, customizable service profiles and a patented Memory Extension Technology for virtual machine scaling.

"It's a platform for a range of applications and a range of workloads," said David Lawler, Cisco's vice president of product marketing for the Server Access and Virtualization Group.

When Cisco pulled the curtain off of UCS in March, details were scant, mainly because the system is based on Intel's Nehalem-class Xeon 5500 series of server chips, which wasn't released until March 30. On Thursday, Cisco revealed performance-test results that show UCS performs either first or second in benchmark tests against competing systems in trials conducted by VMmark and SPEC, which will have full results available soon.

"Although Cisco is a new entrant in the compute space, UCS is performing as advertised," said Jackie Ross, vice president of marketing for Cisco's Server Access and Virtualization Group. "It is delivering the performance."

Cisco also added new details around its Memory Extension Technology, a core component of UCS, which Cisco said enables the CPU to access four times the amount of memory compared to typical blade systems. According to Ross, Cisco's memory extension can cut memory costs by 33 percent to 60 percent in 64-GB, 96-GB and 144-GB deployments, while expanding available memory to 192 GB and 384 GB.

Lawler said the memory boost solves the problem of users running out of memory before running out of CPU availability. Essentially, each DiMM, or unit of memory, the CPU looks for is quadrupled, ultimately cutting memory costs to $3,200 in 64-GB configurations or, in the case of 384-GB capacity, to $47,520.

Cisco Vice President of Marketing for the Server Access and Virtualization Group Soni Jiandani, added that each UCS component, including blades, CPU, disk, memory and I/O cards, are competitively priced against other server vendors' components, but UCS costs about 10 percent less in a standard configuration.

In addition, Cisco said, the UCS system in a 320-blade configuration can dramatically reduce the cost of a full blade system infrastructure by roughly two-thirds. In a 320-blade system, Cisco UCS clocks in at $505,902, while a similarly configured competing legacy system runs roughly $1.6 million.

In that 320-blade system, Ross said, the capital expenses are roughly $12 million, a 40 percent savings compared to similar systems, and cooling and power costs are about $650,000 for three years, a discount of about 19 percent. Ross said Cisco's UCS requires 480 cables and 12 racks in a 320-server configuration, while competing legacy systems would require 3,520 cables and 31 racks.

The majority of the savings is achieved because Cisco UCS does not require a software management license, since the management capabilities are baked in, Ross said. UCS can reduce the number of Ethernet and Fibre Channel switches required by legacy systems by roughly 10 times, instead requiring one Fabric Interconnect per system and one Fabric Extender per system.

Cisco also offered a look at its own 1 megawatt, 10,000-square-foot greenfield UCS deployment, which saved 40 percent cost savings in cabling, fiber, patch cords and labor; had 30 percent more power available to servers; was able to increase the number of physical servers by more than 50 percent in the same space; and could run up to 28,000 virtual machines vs. 7,200 virtual machines in a legacy environment of the same size. That deployment enables up to four times more virtual machines per kilowatt of power, with a minimum of 76 virtual machines deployed per kilowatt of power.

Lastly, Cisco set out to debunk several "myths" of UCS, pointing out that it doesn't require a forklift upgrade; that it can work in environments that are not virtualized; that the pricing is on par, or better, than competing systems; and that UCS is designed to scale.

"The end goal ... is to bring the market a very balanced system," Jiandani said.