Cisco eats own cloud food
Server wannabe Cisco Systems is trying to drive server and storage network convergence and benefit from the inevitable shift to virtualised servers, storage and networks. And the best way to sell a new architecture is to use it.
It was this shift towards virtualisation that compelled Cisco to take on its partners and enter the server racket in the first place nearly two years ago.
Ben Gibson, vice president of virtualisation and data centre products at Cisco Systems, took the stage at the Interop networking - and now cloud computing - trade show in New York to talk about the evolution of vertically integrated, cloudy infrastructure. Gibson's talk covered servers, storage, networking and hypervisors, such as those being sold by Cisco and its Acadia partners, EMC and VMware. There was a bit of the sales pitch, as you might expect, but Gibson also talked about Cisco's own data centre and how it has evolved to the benefit of the networking giant's ability to innovate in its own IT department.
The VBlocks were announced in November 2009 and consist of Cisco rack and blade servers; converged 10 Gigabit ethernet switches; VMware hypervisors and BladeLogic application management tools; and EMC storage arrays, all pre-integrated and tuned up.
HP peddles its own converged stacks, called BladeSystem Matrix, while Oracle has specialised appliances called Exadata for OLTP and data warehousing and Exalogic for web application serving. IBM, meanwhile, is peddling Power and x64 setups called CloudBursts for generic virtualised workloads and Smart Analytic Systems and PureScale machines for data-warehousing and parallel OLTP processing.
First off, Gibson wanted to be clear that Cisco believed there would be best-of-breed engineering in each sub-component of its VBlock stacks as well as in similar stacks coming from Oracle, IBM, Hewlett-Packard and others. The important thing, according to Gibson, is to give customers tight integration and preconfigured systems while also giving them the option of swapping out components here and there.
"We know there are a lot of companies saying they can bring you the whole stack, from application to disk," Gibson said, in an obvious dig at Oracle. "But customers are concerned about lock-in."
What they ought to be concerned about is the premium that vendors will charge for choices outside of their stacks. Who is going to pay for the integration of a NetApp disk array into a VBlock? Hopefully, it was NetApp, but this cost gets shifted onto the customer - in the same way as paying someone do to systems integration (either internally or from outside service experts) has done in the past. The cost doesn't really go away, it just shifts. The best you can hope for with an integrated stack is that the preferred stack is inherently cheaper, because one vendor or a group of partners are collaborating well in development, manufacturing, and support. Even Gibson concedes that the value of an integrated stack is not in features, but in having a single point of contact for the design, service and support as well the stripping out of systems integration costs.
Cisco doesn't just sell cloudy infrastructure, it lives it, says Gibson. A few years back, Cisco's data centres were filled with physical servers without any virtualisation. The Cisco IT department reckons that it spent about 70 per cent of its IT budget on maintaining that infrastructure and the remaining 30 per cent on actually developing new applications to support the networking giant and help it expand into new markets. It took Cisco somewhere between six and eight weeks to deploy a physical server.
Fastforward a few years and Cisco was a little way down the server virtualisation road and now able to get about 60 per cent of its servers virtualised. By doing so, it was able to reduce its IT maintenance spending to 60 per cent of the total IT budget. The time that it took to get a new server allocated to support new workloads fell to somewhere between two and three weeks.
Today, Cisco has deployed its own "California" Unified Computing System blade and rack servers in its data centres and has standardised its application platforms and procedures for configuring and managing its software stack. This means that only 40 per cent of the IT budget is going towards acquiring and maintaining the IT platforms. The remaining 60 per cent is going to new application development. The company has 65 per cent of its servers virtualised and it takes 15 minutes or less to put a new virtual box out there for developers to put apps upon. ®