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

Cisco finally offers Unified Computing System details

Cisco's Unified Computing System includes virtualization-friendly features, but observers wonder whether UCS can set Cisco apart in an already crowded server market.

Cisco finally revealed details about its Unified Computing System's tech specs, cost, and features that could help set the system apart in a crowded server market.

Cisco Systems Inc.'s Unified Computing System (UCS) has been touted as an ideal platform for virtualization because Cisco focused on reducing virtualization pain points surrounding I/O, CPU and memory. But some observers question whether Cisco's decision to enter an already competitive server market makes sense.

For more on Cisco blade servers:
Cisco's planned server play sets up vendor cage match

IT pros ponder Cisco's Unified Computing System

Virtualization users often run out of memory before they exceed CPU power, and it affects how many virtual machines (VMs) can be supported on a single system. Cisco says it solved this issue by developing a patented "memory extension module" that works with the Intel Xeon 5500 Nehalem chip to provide more memory than legacy systems, said Paul Durzan, a systems director at Cisco.

"We invested in this memory expansion technology to increase the number of DIMMs [dual inline memory modules] the CPU can access so people don't have to buy big four-socket servers to get the number of DIMMs they need," Durzan said.

Cisco's USC features, tech specs, and price
Cisco offers two models of blade servers, one with and one without the memory extension technology. The USC B200 M1 Blade Server is a half-width system that supports up to 12 DIMMs and up to 96 GB of memory; up to eight B200 M1s can fit in a chassis.

There is no question that the server business is a low-margin business.
Gia McNutt,

The B250 M1 Extended Memory Blade Servers are full-width, two socket systems that fit four to a chassis. This blade has a whopping 48 DIMMs, but the memory extender technology creates a scenario where the CPU sees four standard DDR3 DIMMs as just one,* according to David Lawler of Cisco's Server Access Virtualization Group. This blade has up to 384 GB of memory.

"The processor can [only] address a certain number of DIMMs, but Cisco puts something between the actual memory and what the processor thinks it is talking to..." Lawler said.

Cisco had not explained exactly how the technology works.

Today, many blade servers have eight or more DIMM slots. For instance, HP's latest blade, the ProLiant BL490c server, can support up to 18 DIMMs and up to 144 GB of memory per blade using registered DIMMs -- which is more than Cisco's B200 M1 Blade Server.

HP plans to launch its own unified system called the Matrix on April 20 that will compete with Cisco's UCS, according to the Wall Street Journal.

Another feature Cisco hopes will set UCS apart is a virtualization adapter embedded into the system called VN-Link, which lets VMs bypass the hypervisor and talk to one another directly. It also lets users manage VMs through the adapter in the same way the physical network is managed, and users can set policies that follow the VMs, according to Cisco.

Due out later this quarter, the system consists of a 6U chassis (UCS 5108) and up to eight 1.25 inch wide B-Series blade servers in a standard 19-inch rack, or four full-width blade servers.

UCS also supports up to two fabric extenders and uses standard DDR-3 memory, industry-standard disks and standard Ethernet connections. Users don't have to use Cisco switches, but Cisco recommends them. Cisco is finalizing the power consumption data, so no information is available. But the company did reveal that it will be based on AC and possibly DC power.

As for networking, Cisco's B-250 M1 Extended Memory Blade Servers support two dual-port mezzanine cards for up to 40 Gbps of I/O per server. This, Cisco said, prevents the bottlenecks that are common in virtual environments. Users can also scale the bandwidth to deliver up to 10 Gbps of I/O bandwidth to an individual server blade.

UCS Management, cooling
The system also includes embedded management software that allows IT pros to manage the pool of blades and networking components within a system from a single console. It is set up so that server, storage and network administrators all have their own section of the graphical user interface in which to do their work. There are no licensing costs associated with the management software, Cisco reported.

Another feature is called "service profiling," which allows admins to assign configuration settings to each individual blade, giving each blade its own "personality." This is beneficial when moving a failed blade or upgrading a server; the new blade will assume the same settings as the blade it replaces. Administrators can also copy the "identity" of a particular blade and put it on a new blade to scale up quickly, Durzan said.

As for cooling, the backplane of the chassis is 63% open, which allows front-to-back cooling, Durzan said. "Having a backplane that is 63% open is huge; most blade chassis are only about 20% open in the back," Durzan said. "This way, we can also handle higher-power CPUs as they come out in the future without worrying about too much heat."

Cisco offered pricing for UCS, but only in terms of a system with 320 blades, because large installations are where Cisco thinks UCS will have the most value. Cisco broke the cost down as follows: $1,812,800 for the servers and $239,000 for the adapters, $167,880 for the chassis, $138,182 for the fabric interconnect, and $159,920 for the Fabric Extender, for an overall total of $2,518,462. Taken together, a fully loaded UCS with all the necessary adapters and fabric interconnects costs $7,870 per blade.

Cisco said they will not have any customers to talk about the system until the fall.

IT community unsure of Cisco's UCS
Gia McNutt, the CEO and co-founder of the voice and data solutions company SOS, is a Cisco partner that also sells HP, Compaq and Dell servers. She said the server market is already crowded, and she questions Cisco's decision to enter the server space.

"[Cisco] is my favorite manufacturer/partner. And yet I wonder about their wisdom in entering this highly competitive, low-margin server market, even though they have their new focus on data centers," McNutt said.

McNutt said Cisco has put its hat in the server ring because, like many companies, they "feel an irresistible pressure to extend the equity of their brand."

"If a corporation is doing smashingly well in some lines of business, they feel they can enter every market and do well. Mostly, this is not the case," McNutt said, noting that there have been many companies whose extensions fail, or become barely profitable segments of the business.

"There is no question that the server business is a low-margin business. The best in the business struggle for profitability, with a lot of competition. Cisco's own MCS [Media Convergence Servers] platforms are 50% or more expensive than the OEM manufacturers' brands. That's probably because the prices of RAM and disk space and CPUs fluctuate so rapidly, and it is no different in this market," McNutt said.

Others agreed that despite its well-known name, Cisco will find it difficult to make headway in the server market. Despite its well known name, said Jon Oltsik, senior analyst with Enterprise Strategy Group.

"This won't work in the short term, as users will have to manage a mix of server platforms," Oltsik said. "Longer term, the server vendors will certainly address Cisco's packaging with their own integration and potentially industry standards. It won't be easy to push aside Dell, HP, and IBM based on a proprietary design."

*Cisco clarified its memory extension technology after publication. The company initially verified that the memory technology virtualizes memory, but that definition is not 100% accurate, Cisco said.

Let us know what you think about the story; email Bridget Botelho, News Writer.

And check out our data center blogs: Server Farming, Mainframe Propellerhead and Data Center Facilities Pro


Dig Deeper on Server hardware strategy

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.