It only happens once or twice a decade: one of the ‘gorillas’ of IT is drawn to an adjacent market, that it may perceive as an opportunity or a threat or both, and decides it wants a part – a large part – of the action.
It happened in the 1980s when IBM responded to Digital Equipment and Data General by launching itself into the minicomputer market with the System/3x range, and again in that decade when it took a look at the scores of emerging personal computer companies and created the IBM PC. It happened when Oracle decided it had outgrown the database market and started an expansion into business applications. It happened in the 1990s when Microsoft moved beyond the confines of the desktop software market and threw itself into developing the server operating system Windows NT.
Cisco – the $35 billion market leader in networking systems – thinks that its Unified Computing Systems (UCS) is just such a move. And by many accounts, as with similar moves in the past, its entry into what is today narrowly defined as ‘the server market’ but which is probably more accurately termed the data centre systems market is set to have a profound impact on both customer engagement and the IT industry’s competitive landscape.
Certainly, the initiative is much more than just a ‘me-too’ play in blade servers. Although it has built up extensive experience of selling networking and storage switches into data centres, when adding a compute environment to that line up Cisco has had the luxury of being able to design an architecture from the ground up. And the architecture has been built with a completely new set of customer imperatives in mind, ones that put the virtual machine at the centre of the data centre universe.
So rather than a latecomer to the server market, Cisco – and others with it – see its move as creating an early-mover advantage in the virtualising world of data centres and cloud computing.
As such, UCS will make Cisco as many enemies as friends. The launch of a mainstream data centre computing platform, that provides a unified fabric for input/output (I/O) across server, storage and networking systems, sets Cisco on a collision course with the world’s largest systems vendors (HP, IBM, Dell, Fujitsu and Oracle/Sun) by offering medium and large enterprises a single architecture that “links all data centre resources together”, overcoming the “assembly-required” nature of today’s distinct virtualisation environments.
Project California, as UCS was known through most of its three-year development, is “one of the largest endeavours ever attempted by Cisco,” observes Silvano Gai, a renowned Cisco engineer and co-author of a book documenting the scope of the project.
And the opportunity for that undertaking presented itself because one single factor – virtualisation – was changing the game.
“We recognised that virtualisation would create an industry transformation,” says Jim DeHaven, head of Data Centre Solutions for Cisco UK and Ireland. “That is where the real focus was when [we were] considering entering an adjacent market: the understanding that the building block was no longer the server hardware but the virtual machine. [And that UCS] would be a new architecture in support of this new atomic building block. That is what would give us differentiation.”
That change also allowed Cisco to leverage its heritage. “California is a scalable compute platform based on the natural aggregation point that exists in the data centre: the network,” says Silvano Gai and co-authors. As such, it converges the core fabric that underpins the servers, storage and networking into a single network, with unified I/O, and in doing so it consolidates several of the components that support those traditionally distinct environments.
“Only Cisco could do that: effectively make a [virtual] compute node a member of the network in a complete way,” argues Jay Kidd, chief marketing officer at storage systems vendor NetApp (a partner on UCS), “being able to virtualise the network address, the quality of service parameters and all the things that are truly network-dependent.”
And even in the build up to the general availability of UCS in early July, Cisco is promising that clients will reap huge benefit from such an integrated, network-centric fabric. UCS “reduces total cost of ownership – up to 20% reduction in capital expenditure and up to 30% reduction in operational expenditures,” it claims.
Others who have worked closely with Cisco on the project think it is being conservative in its estimates, especially in areas such as virtualisation management. “In our studies, we’re seeing a 80% to 90% reduction in many of the provisioning, patching and configuration management tasks,” says Jim Grant, senior VP for corporate strategy and development at service management software company BMC, whose software is bundled with UCS.
That has been achieved, Cisco says, by a whole series of innovations that were largely developed at the Cisco ‘spin-in’ Nuova Systems (see box: ‘Acquisition No. 126’), where a group former Cisco executives worked for three years in secrecy.
It was a unique opportunity, says Soni Jiandani, the former chief marketing officer at Nuova and now VP Marketing for the Server Access Virtualisation business unit at Cisco. “We designed this from the ground up, from the silicon to the system level,” she says.
The authors of Project California identify five key areas of innovation:
• Unified fabric: California is the first server designed around the concept of a unified fabric. “This is important since different applications have different I/O requirements. Without [that] it is difficult to move applications from one server to another while also preserving the appropriate I/O requirement”. As Soni Jiandani at Cisco points out, “virtualisation is forcing the different silos of the data centre – storage, compute and networking – to work together.”
• Embedded management: UCS has an on-board management system that has global visibility into all elements that constitute the system devices – whether physical or virtual – and their configuration. Through Unified System Management, management is integrated into all the components of the system, enabling the entire solution to be managed as a single entity. That allows data centre managers to provision applications in minutes instead of days.
• Optimised virtualised environment: UCS is designed to support a large number of virtual machines not only from the point of view of scale, but also for policy enforcement, mobility, control and diagnostics.
• Fewer servers with more memory: UCS computing blades use new Cisco Memory Expansion Technology to support up to four times more memory than is available on rival servers with the same processors. More virtual machines per server also means lower power and cooling per virtual machine.
• Stateless computing: Critically, in UCS server attributes are not tied to physical hardware, enabling seamless mobility. The core idea behind ‘stateless computing’ is the ability to dynamically provision resources: “Every application workload is different, therefore it requires different levels of computing resources – storage, memory, I/O, network connectivity – all of those things have to be managed together so they mesh like gears,” says Jim Grant at BMC.
That – plus several other innovations – puts UCS on an elevated level, according to Grant. “People have said this is just another blade server. That is so not it, believe me. Think about it: if you had a blank slate, and wanted to design computing in a virtualising world, you would have to take in lots of things, but an integrated networking capability and a highly integrated storage access capability would be key elements.”
He continues: “Cisco has discovered all the main bottlenecks in virtualisation and attacked them. For example, the problem of insufficient memory to handle the huge transfers of operating system and application stacks. What did they do? They created an application specific integrated circuit (ASIC) which can triple the memory capability.” (Cisco says quadruple.)
Another of the bottlenecks it addresses is ‘I/O to storage’ that comes about when large systems images are being moved to and from storage. “Being able to boot a computer from a remote storage device eliminates the bottleneck there,” says Grant.
Of course, UCS does not have all the answers: Critical to Cisco’s challenge will be its ability to draw on the expertise of key partners. Its compute capabilities, UCS B-Series blades which offer systems of up to 320 compute nodes housed in 40 chassis with data flowing across 10 gigabit Ethernet, will be based on Intel Nehalem processors, the follow-on generation from Intel Xeon; VMware will supply the critical virtualisation software; BMC is behind the provisioning and configuration management software; EMC and NetApp will be responsible for the storage system units; Emulex and Qlogic will input storage networking technology; Oracle will deliver middleware and systems software will come from Microsoft and Red Hat.
Slicing into costs
That ecosystem – and it is growing – is gathering around UCS because they believe it will have a major impact on some of the well-documented problems of today’s server environments: Low utilisation rates, high-power consumption, cooling issues, and management complexity.
An overriding aim with UCS was to eliminate or reduce some of the unnecessary waste associated with the integration of separate and functionally duplicated storage, server and networking components. This lowers data centre costs by cutting the required number of network adapters, switches, and cables and by decreasing the necessary power and cooling.
In all, Cisco claims UCS involves one-third fewer components – switches, adapters, management modules, servers – as well as reduced points of management.
Early benchmarks for UCS, run by Cisco in a 1 megawatt, 10,000 square foot green field facility, are (at least on paper) impressive:
• 40% cost savings in cabling, fibre, patch cords and related labour. In a deployment of 320 physical servers, 480 cables were used, as opposed to 3,520 required for a ‘legacy’ system
• With the help of extended memory, the system could run up to 28,000 virtual machines, versus 7,200 in a legacy environment of the same size
• With the extended memory enabling a greater concentration of virtual machines per server, the footprint of the total 320 server environment was cut from 31 racks to 12
• The system could sustain up to 4 times more virtual machines per kilowatt of power, with a minimum of 76 virtual machines being deployed per kilowatt of power
• For a 320 server system, $650,000 would be spent on power and cooling over 3 years, versus $800,000 at a similar legacy environment, an almost 20% reduction.
As more independent benchmarks become available in coming months, UCS’s differentiation will be confirmed, says its supporters. Maybe then, critics who deem it “another blade server” will be silenced, says Douglas Gourlay, vice president of Data Center Solutions at Cisco, in his blog: “With something that is this big in scope and scale, people often have a hard time wrapping their heads around it and conceptualising it. In a couple of years, they [may] look back at the piecemeal, service-intensive, high-cost approaches they took in 1990-2010 and wonder why no one ever did Unified Computer earlier.”
Indeed, Sun Microsystems may be famed for stating – well ahead of its ability to deliver – that ‘the network is the computer’. If informed insiders are right, it looks like it took a networking company to prove it.
READ MORE ON CISCO’S UCS
The stealthy ‘spin-in’ behind Cisco’s data centre play
Universal Computing System was developed by four Cisco executives in a satellite company called Nuovo Systems