Cisco UCS may be winning the hearts and minds of enterprise data center managers, but service providers say the...
proprietary system isn’t a good fit for the cloud.
Last week, Cisco Systems Inc. said it had logged the 11,000th customer for its 3-year old Unified Computing System (UCS), was positioned Gartner’s “Leaders Quadrant” for blade systems and launched the third-generation of the converged infrastructure platform.
Meanwhile, some cloud and hosting providers say they are moving off UCS.
One cloud hosting provider and UCS early adopter that went live with Cisco’s gear in 2009 started taking it out its data center just one year in to the relationship, the company’s CTO told SearchDataCenter.com.
One problem is that Cisco UCS costs easily 30% more than comparable “enterprise commodity” gear from vendors like Hewlett Packard and Dell. The CTO, who requested anonymity to preserve a relationship with Cisco on the networking side, wasn’t impressed with the UCS Manager and Service Profiles functionality either.
Cisco UCS originally appealed to the firm “because it was a single integrated system,” but it quickly came to the conclusion its configuration – UCS plus NetApp storage – “wasn’t very flexible,” he said.
Flexibility is key for service providers, said Christian Teeft, vice president of engineering at Latisys, a data center colocation and hosting provider in Englewood, Colo., that is launching a cloud service. Teeft has evaluated UCS but uses HP rackmount servers. Teeft said he chose traditional rackmount servers to avoid lock-in.
“With UCS, the challenge is that even though the servers might be amazing, you’re tied in to Nexus [Cisco’s switching platform],” said Teeft. Once you’re locked in, you lose leverage in terms of price negotiation, he said, and the freedom to look around for the latest and greatest new technologies.
[Ed. note: In fact, it is technically possible to connect Cisco UCS Fabric Interconnects to non-Cisco switching gear, but Cisco does not encourage those deployments, nor is it a common practice.]
UCS doesn’t add up
These comments follow on news that another Cisco UCS service provider shop, Harris Corp., is exiting the cloud hosting business and is liquidating its data center assets for a loss of $70 to $80 million on a $200 million investment. Harris was a poster child for the Virtual Computing Environment Company’s (VCE) Vblock, a hardware suite that combines Cisco UCS and Nexus compute and networking, EMC storage and VMware virtualization.
The news that cloud service providers are having trouble monetizing their Cisco UCS and VCE investments does not surprise Greg Ferro, a blogger and network architect focused on Cisco and data center networking.
“Service providers just can’t make it pay with Vblocks,” he said. A low-end Vblock can easily cost $1.5 million, but more often than not can cost $5 million, he said. Cloud hosting, meanwhile, is a very competitive market, faced with constant downward pricing pressure, making it hard to turn a profit.
To stick with UCS would have meant charging its customers four times more than it currently does to maintain its margins, said the cloud provider CTO. That’s an untenable proposition for cloud providers, considering Amazon Web Services continues to drop its prices. “We cater to the enterprise, which isn’t quite as price-sensitive, but we’re still driving toward zero,” he said.
Cisco UCS apt for enterprise data centers
Service providers are quick to point out that even though UCS doesn’t work for them, it can still be a great fit for enterprise data centers that could benefit from the throughput and the extended memory UCS provides, said Teeft.
“I do think that enterprise IT organizations can leverage UCS and get some value out of it – in terms of the number of partners that they have to use, and the tools that can simplify deployments,” he said.
And shops that do buy UCS have plenty of new goodies to choose from. Piggybacking on the release of Intel Corp.’s latest Xeon E5-2600 “Romley” processors, Cisco last week announced three new “M3” UCS blade and rackmount servers equipped with the Intel chips, a new chassis I/O module that offers up to 160Gbps of bandwidth down to the chassis and a new Fabric Interconnect that doubles UCS switching fabric to just under 2Tbps.
Cisco UCS Manager now also manages both blade and rack servers, where it was once limited to blades. Furthermore, in the second half of this year Cisco will introduce centralized management that supports thousands of UCS blades spread across multiple UCS domains across geographically distributed data centers.
That last piece of functionality is particularly useful to service providers, said Ken Owens, vice president of security and virtualization technologies at Savvis Inc., a cloud and managed services provider that uses UCS as well as other platforms. “The multi-domain feature gives us the ability to control our [Disaster Recovery] capabilities across regions,” he said.
At the same time, Savvis doesn’t rely on Cisco’s management layer per se, but writes its own management software that taps in to low-level APIs provided by Cisco to write software that can span a variety of environments – some that are less feature-rich than UCS.
“With cloud, we’re also looking at not taking advantage of new technologies to keep our costs down,” Owens said.
No service providers, no problem?
Cisco declined to break out the percentage of its UCS business that goes to service providers but claims it has made “a huge investment” in the server provider space. Despite some service providers moving away from UCS, Cisco claims it’s having its best success with service providers, especially those that value its integration work across component types, said Brian Schwarz, Cisco director of product management for UCS.
It provides a good alternative to building, integrating and maintaining all of your own systems, according to Schwarz.
From prior experience working at the now-defunct Exodus Communications, the Internet service provider that famously burst with the dot-com bubble, Schwarz cautioned service providers against attempting to lure enterprise customers with commodity systems.
He believes IT hosting providers should mirror customers’ internal data center as closely as possible
Of course, Cisco has a business reason for pushing UCS over commodity hardware. Some say that in the era of cloud and virtualization, the brand of hardware no longer matters.
“One thing that virtualization has done is that no one really cares about the hardware anymore,” said the CTO that recently eliminated its UCS gear.
“If you look around, [proprietary platforms] just aren’t the way things are headed. Cloud computing is going commodity,” he added.
Even if no service provider bought another UCS ever again, Cisco probably wouldn’t suffer much; it has a strong foothold in the enterprise.
Enterprises currently buy “the lion’s share” of servers overall, said Jed Scaramella, research manager at IDC Corp. and only about a third of enterprise IT budgets are directed toward hosted, outsourced or public cloud, with the remainder focused on traditional IT and private cloud.
But a shift to the cloud could change that.
“If the enterprise all goes to Amazon EC2 and Rackpace, then that’s a problem for all branded systems vendors, because their strategies aren’t 100% aligned with service providers,” said Jonathan Eunice, principal IT advisor at Illuminata Inc., an IT consulting firm in Nashua, N.H. “They’ll have to go back to the drawing board.”