Cisco Systems has been one of its best beta customers for its widely
hyped Unified Computing System data center technology, deploying the
technology in one of its own data centers to run several applications,
according to company officials.
Cisco also plans to replace the bulk of the x86 servers currently in
their 52 data centers worldwide with the UCS technology—codenamed
“California”—over the next two years, according to the officials.
The UCS, announced by Cisco in March, is an all-in-one offering that
brings together computing, networking, storage and software management
into a single offering. It includes not only Cisco server and
networking products, but also offerings from partners such as VMware
and its vSphere platform for virtualization, EMC for storage and BMC
Software for management software.
Click here to see the building blocks for Cisco's UCS.
The product, which is not yet shipping, includes Cisco blade servers powered by Intel’s quad-core Xeon 5500 Series “Nehalem EP” processors and
networking technologies such as FCoE (Fibre Channel over Ethernet) and
10 Gigabit Ethernet. Cisco offers the FCoE fabric in its Nexus 7000
switches.
Cisco’s strategy is part of an overall trend in the data
center—fueled by such technologies as virtualization, cloud computing
and Web 2.0 applications—to converge resources. For example,
Hewlett-Packard in April rolled out such an offering with its BladeSystem Matrix that integrates server, storage, networking and software into a single product.
Speaking on a Webcast May 27, two Cisco data center officials said
their goal is not only to incorporate the UCS into their own data
centers, but—like any beta tester—to use their experiences with the
technology to give informed feedback to Cisco’s engineers.
The UCS will be a key building block for their own data centers
going forward, enabling them to scale their infrastructures while
driving down power and operating costs, they said.
“We’ve very much integrated it into our current models,” said Chris
Hynes, director of IS for Cisco’s Network and Data Center Services
group.
Cisco has been running the UCS product in its Mountain View, Calif.,
data center since mid-March, before the company announced the offering.
The systems currently are running the site for Chairman and CEO John
Chambers, Cisco’s public relations site and a host of legal and
financial applications, according to John Manville, vice president of
IT at Cisco.
The UCS product is running a host of virtual machines as well as an Oracle database that has not been virtualized.
Cisco also is using a new 10,000-square-foot data center it’s
planning as a platform for UCS, Hynes said. The 1 megawatt facility
will run UCS boxes, saving the company money in a number of areas. A
traditional data center of that size would need 135 server racks, with
4,320 Ethernet cables and 2,160 copper cables, he said. With the UCS,
all that will be reduced to 72 racks, 1,008 Ethernet links and 300
copper links.
Hynes said the reduction in cabling is important because it frees up
space and resources to add more computing power to the data center. It
also saves money, he said. Cabling in the traditional data center would
cost $2.7 million; with the UCS, that drops to $1.6 million, he said.
The number of physical servers that can fit into facility grows from
720 hosting up to 7,500 virtual machines in the traditional data center
to as many as 1,400 servers hosting 12,000 to 14,000 VMs with the UCS
technology.
Virtualization technology will continue to be a key part of what
Cisco has planned for all of its data centers, which cover about
215,000 square feet throughout the 52 facilities, Manville said.
Currently, about 30 percent of all the servers are virtualized, he
said. Over the next two years, the goal is to increase that to 70 to 80
percent.
With all these capabilities coming into the data centers, Cisco
officials also are in the process of creating an internal cloud
computing environment which will be based in large part on UCS. Dubbed
CITIES—for Cisco IT Elastic Infrastructure Services—the goal of the
project is to bring as many applications as possible onto the internal
cloud, Manville said.
Eventually Cisco will move to a hybrid model, moving applications
and workloads between the internal cloud and public clouds depending on
need, he said. The first phase—called “Mist”—will launch in August or
September, he said.