Cisco's Unified Computing System
(click for larger image)
Cisco is making a bet that it can master the art of blade manufacturing faster than blade makers, such as IBM and Hewlett-Packard, can move onto its switching turf. If it's right, the dominant networking vendor will have opened a major new market for itself and provided blade users with a new type of server, the Cisco Unified Computing System.
Even if it's wrong, the move is sure to shake up accepted blade designs and proliferate options, to the benefit of end users. Blades, a new form factor less than a decade old, now make up 20% of the server market, according to IDC. At stake is the possibility raised by Cisco that revamped blades -- one of the most profitable segments in servers -- can be combined with converged network components, then linked to virtualized network and storage resources under a single management interface.
Such a move promises to drive costs out of the data center by prompting more efficient utilization of devices at every level. It would also allow the benefits of virtualization to move beyond server consolidation into overall management of the data center as sets of pooled resources -- servers, networking, and storage -- potentially reducing IT operating costs. Cisco said the savings could amount to 35%.
But to accomplish all these goals, Cisco will have to bring a new type of blade to market, one that relies on a 10 Gigabit Ethernet fabric to feed I/O to both storage and networking devices. The idea isn't completely new. Since January HP has offered BladeSystem with Virtual Connect Flex-10, which divides overall I/O between the enterprise network and iSCSI storage. Gary Thome, director of strategy for HP's blade and infrastructure software unit, said HP customers won't have to wait for a new management interface to manage the two; it has one today in Insight Dynamics VSE. He claims Cisco is the "plumber" of network pipes, but HP is "the general contractor to build the whole house. We don't see the data center as a network with servers hanging off the end."
Burton Group analyst Nik Simpson said several blade manufacturers have today or soon will have the capacity to combine network traffic and storage data as a single stream coming off the blade to networking devices on the blade chassis. Cisco "incrementally goes a little further" in its ability to achieve the convergence."
An incremental step may be enough to give Cisco a foothold.
"Nobody of Cisco's size has come into the [blade] server market since Dell" did in 2001, Simpson said. Despite its size, Cisco will still have to convince server managers, who aren't necessarily in thrall to the network managers that Cisco normally talks to, to buy into a new line of blades. It will have to continue to invest in the design and manufacturing of blades, something that HP, Dell, IBM, and Sun have worked on for years. "Cisco has never moved into a market this size with so many established incumbents," Simpson added.
Cisco has assigned a key role in the process to BMC Software, the Houston systems management vendor, to deliver a combined management tool to supervise the different virtualized resources. Getting one management interface for servers, networking, and storage has been a stumbling block in the past. The fractious storage industry hasn't come up with a single interface for storage, because of its many competing vendors and conflicting standards. Several players have tried to advance a unified approach to deploying storage resources, said Simpson, and the industry "is littered with their skeletons."
This time around, virtualization vendors Microsoft and VMware may have the heft to overcome those barriers. Both Microsoft and VMware have the muscle to lay down a storage API for virtualized systems and insist storage vendors link their devices to servers through it, said Simpson. Such a unified storage pool is needed to realize the benefits of virtualization. Virtual machines to be moved from physical server to physical server need a single storage file system underneath them. VMware has gotten around the fractious industry barrier by coming up with its own. It's promising a storage API as part of its vCloud initiative that will let any storage vendor adopting the API to participate.
Beyond the storage, there's still the issue of delivering a single management interface that cuts across servers, storage, and networking. Cisco has vested its new partner BMC with the task of delivering a tool that understands converged network communications on a 10 Gigabit Ethernet fabric. It will understand how to divide that traffic across pooled networking and storage devices and how to divide up the capacity of existing devices between competing demands.
Existing management tools often go through a discovery and mapping phase, where they log the activity on various devices and gain some metrics for predicting traffic. The tool must then present the manager with choices on how to divide up his resources, and sound alerts when those choices are about to go awry.
When it comes to virtualized resources, however, the problem is compounded by the tendency to overallocate servers with resources. Since future demand can't be completely predicted, server administrators guard against failure by overprovisioning with CPU cycles, memory, and storage. To balance workloads and avoid overprovisioning, Cisco will need a new management tool that can view each of these virtualized resources and then apply the smarts to manage them as part of a single system.
Simpson said BMC and CA are the systems management vendors that are independent of IBM and HP, also systems management vendors, and in a position capable of delivering on a cross-vendor, total data center view of virtualized resources. Cisco chose BMC. Simpson said "don't hold your breath" that the tool will meet all the demands upon it in its first issue.
BMC CEO Bob Beauchamp is an enthusiastic champion of Cisco's move, and said BMC will meet the goals laid out for it. The Cisco blade "is a game changer; they've cracked the code on several bottlenecks."
BMC's flagship product, Patrol, was an early in adding cross-vendor virtualization. Now it will try to graft on cross-resource virtualization. "The pain point," he concedes, "is where virtual machines and the virtualized network meet up with storage."
If Cisco produces standard Intel-based blades with a unified management tool assigning them VMs, allocating and reallocating their resources based on need, and is able to move them around physical servers atop a virtualized storage system, then Cisco will have simplified data center operations. Instead of buying a server that fits an application, IT managers will buy standardized blades and run each application in a VM that matches the application's needs, perhaps on an hour-by-hour basis.
"What we're bringing to the table is a top down view" and the ability to provision a small virtual machine, say, for running Oracle in a department versus a large one for running Oracle at the enterprise level, Beauchamp said.
The manager of the virtualized data center "shouldn't have to worry how many network interface cards are on the server or how much memory is available," said Beauchamp. Indeed, the Cisco blade design is aimed at making such configuration problems an issue of the past. Instead, the administrator using a cross data center tool will point and click on the resources he needs, and the underlying system will "make it so," he said.
Want to hear more about server virtualization and data center consolidation? Attend a Webcast on this topic on March 31. Find out more. (registration required).