check => 2 LG =2; LGCountry=2
Versichern Sie Ihre Stromversorgung mit NO-BREAK KS®
Media EURO-DIESEL in the news Cisco says new data center roadmap works | 2010.06.10

Cisco says new data center roadmap works and has numbers to prove it

10 Jun 2010 | DCD Focus E-Newsletter
by Yevgeniy Sverdlik



Cisco Data Center Racks

Company reports progress on CITEIS and new Texas data center; offers TCO figures on architecture changes made so far.

Cisco, the San Jose, Calif.-based IT colossus, prides itself on using its own technology in its data centers. About two years ago, the company started a complete overhaul of its approach to enterprise IT infrastructure and today, half-way through the new roadmap, the strategy seems to be paying off. And the strategy, in a big way, involves the company’s Unified Computing System – the block of servers and networking equipment the company launched a little longer than one year ago.


There is a number of challenges Cisco’s IT organization is currently faced with and they are coming from multiple directions within the enterprise. Business-side challenges put pressure on the organization to be more productive and to improve SLA’s. Cisco is constantly entering new businesses and with those businesses come new service-level requirements.


On the operations front, there is pressure associated with finiteness of power and cooling capacity and pressure to increase asset utilization.

One of the key drivers behind the company’s current infrastructure strategy is higher CPU utilization rates.

“If you do increase utilization to whatever level you think is appropriate for your company, you can build less data centers (or) defer having to build them in the future,” said John Manville, Cisco’s VP of IT and Network and Data Center Services, in a presentation in June.


There is also pressure to speed provisioning time for new IT capacity and – of course – to become “greener.”



The company started revamping its data center strategy about two years ago. “Cisco has identified a roadmap for how data centers are evolving,” Manville said. “We are well on the way down this roadmap.”


Key tenets of the new approach are running IT as a business – acting as an internal service provider – and building a service-oriented architecture that maximizes the use of virtualization.


The progression of steps on this roadmap is as follows: consolidation, virtualization, automation, utility and market. According to its VP of IT, Cisco is currently in the automation phase and will very soon be in the utility phase, meaning IT services will be delivered as a utility, self-provisioned, with capacity based on needs. The final market phase is when the architecture will make use of public service providers to move services wherever is most appropriate at the time, as market forces and business requirements dictate.

Cisco Data Center


In terms of total cost of ownership, the strategy has been paying off and Cisco expects to continue on the same path. After the initial step of moving from a legacy, all-physical architecture to a 60 percent virtualized one, the average compute TCO dropped by 37 percent: from more than $3,500 per OS instance per quarter to less than $2,500. The organization’s delivery time improved as well, going from between six weeks and eight weeks to between two weeks and three weeks.


Adding automation into the mix improved the TCO by another 16 percent. In the architecture’s current state, Cisco’s average TCO per instance per quarter is less than $2,000. It also now takes about 15 minutes to provision a virtual instance.


The organization’s next step is to transition to a UCS-supported internal cloud, where 65 percent of workloads will be virtualized. It projects an additional 17 percent reduction in average TCO. The final stage – with 80 percent virtualization rate – is expected to bring another 24 percent in savings.


Beginning with the next 65 percent-virtualized phase, users will provision VM’s in the cloud themselves, at which point the organization will reach its desired utility-style service.


Costs going into the TCO calculations included everything except storage: data center space, operations, staff, capital expenditures on networking and compute equipment and software licensing.



The company calls its cloud CITEIS, or Cisco IT Elastic Infrastructure Services. CITEIS is delivered in part by Cisco’s UCS, with storage solutions from NetApp and EMC.


Manville’s organization is in the process of moving its existing virtual environments onto UCS, with half of them expected to run on UCS by the end of July. He plans to have all virtual environments running on the Unified Computing platform by the end of July of next year.


Initial implementation of the hybrid cloud – extending CITIES into the public cloud – is expected within the next 12 months. Cisco set up 16 racks filled with the Unified Computing System chassis at VMworld 2009 in San Francisco. Image courtesy of Cisco



The new build-out strategy is based on building the right capacity in the right places and optimizing demand in facilities. Another part is resiliency and the company is working toward moving its critical applications out of earthquake zones, namely the Silicon Valley, where the headquarters and some of its data centers, running critical loads, are located.


CITEIS will live in two data centers in Texas, one in Amsterdam and a number of other facilities around the world. Of the two Texas data centers, one is a 1 MW colocation site and the other (Texas 1) is Cisco’s own 10 MW facility, located in Richardson. The company is in the middle of building a second internal data center (Texas 2) from the ground up in Richardson and the colo will eventually be phased out.


Certain critical applications will be deployed in an active-active configuration between Texas 1 and Texas 2, addressing further the need to increase resiliency. Other applications will use a disaster-recovery facility in Raleigh, N.C., for failover.


A CITEIS proof-of-concept set-up is currently running on 10,000 sq ft in the Texas colocation space. When Texas 2 comes online (early next year) workloads – whenever possible – will be moved virtually to the new facility. The rest will be moved physically with the goal of decommissioning the colo completely, said Bob Scarbrough, member of the Cisco IT technical staff.


Decommissioning the Texas colo is only one step in Cisco’s ongoing data center consolidation program, whereby the company is planning to go from about 50 sites around the world (some are just IT closets) to about 20.



Texas 2 is a completely different design from what Cisco is used to. The 26,000 sq ft facility will have half as many cabinets as Texas 1, while having similar or greater compute capacity, Scarbrough said. While power density in older Cisco data centers is between 3 kW and 5 kW, the new facility will support up to 18 kW per rack. Each rack will be able to hold up to five UCS chassis.


The system will deliver 440 V to each rack, which will then be converted down before it reaches IT equipment. In case the industry does at some point switch to 440 V gear, adapting Texas 1 to the change will simply be a matter of removing the converters, instead of making changes on the wider infrastructure level.


Texas 1 UPS systems have 1,920 batteries. Texas 2 will have zero, relying instead on rotary UPS systems from E1 Dynamics. The new design also uses geothermal wells for energy storage.


The facility has 30 ft ceilings and no raised floor.


This will be the first Cisco data center to use an airside economizer, which Scarbrough expects to generate free cooling up to 50 percent of the year. For the rest of the year, the facility will use its six chillers: four 1,000 ton units and two 500 ton ones. The 500 ton chillers have variable-speed motors for dynamic capacity control through an automated monitoring and management system.


Images courtesy of Cisco