It's all made to sound so easy: first virtualization, then the cloud, and then an IT utopia of unlimited resources and global operations.

The reality, of course, is quite different. As more and more enterprises are finding out, the transition away from physical resources to both virtual and cloud architectures brings with it a host of challenges. But from an infrastructure standpoint, the most significant is ensuring that vital data can get where it needs to go quickly and securely. And for that, there is no substitute for a high-speed network that is both flexible and scalable.

But how do we get there? Or, more precisely, what sort of network infrastructure is most suited to the coming cloud paradigm?

At the moment, the answer to that question is a moving target, which probably won't come to rest until more real-world experience with virtualization and the cloud is in hand. But the fact that more research is shifting in this direction -- particularly now that the cloud is beginning to look like a real movement, not a fad -- indicates that network infrastructures are as ripe for conversion now as servers and storage were at the beginning of the virtualization era.

One of the newest approaches is the concept of the "dynamic infrastructure." Rather than a simple collection of humming boxes or cards designed to push data this way or that, the dynamic infrastructure brings together virtual networking, automation and resource management with tools like application management, security and policy management to create a self-managing environment that can react to changes in workloads and other needs with minimal human interference.

Lori MacVittie, technical marketing management for application services at F5 Networks is one of the prime movers of the concept, which she says will be the inevitable result of the transition to the cloud.

"When the entire data center is founded on a dynamic infrastructure, the infrastructure can react itself to changing network and application conditions and needs," she says. "When the entire ecosystem is sharing status and information about performance, every component can adjust itself dynamically to what’s needed now to improve performance or maintain availability. And it happens automatically, based on the specific needs of the business and IT."

It was virtualization more than anything that demonstrated the increasingly organic nature of the modern data center. Once systems went virtual, it was no longer a simple matter of provisioning new equipment and reaping the benefits of greater overall performance. Now, changes in one area can have a dramatic impact on another in an ever-shifting environment of new data pathways and policies.

That has forced many enterprises, even those that have deployed the latest in virtual I/O technology, to pay strict heed to what is happening on the network level, particularly as it struggles to keep pace with advanced computing technologies like multicore processing.

"Platforms like (Intel's multicore) Nehalem call for a much different virtual I/O solution than is mainly deployed today, simply because these platforms are so extensive and add so much more computing power," says Greg Schere, CTO at Neterion. "I/O has not kept pace with this revolution, so people are virtualizing low-performing applications and things that don't require a lot of I/O. We've fixed the CPU problem and the memory problem -- now we have to fix the I/O problem to scale up with these systems. We need to break the glass ceiling and extend the virtualization value proposition that is being made with all these gains happening on the processor."

At Neterion, the strategy is to focus on technologies like hypervisor offloading, which puts much of the networking load onto the adapter, coupled with embedded QoS capability and Virtual Line Technology that allows a single adapter to appear as multiple ones.

For some, the fact that the CPU quite often does double duty maintaining the hypervisor and the network connectivity is one of the primary reasons most data centers can't put on more than a few virtual machines per server. This vastly undercuts the true potential of current server platforms, which promise consolidation rates of maybe 50:1 or more. That's why more networking solutions are bent on dividing that labor with various offload technologies -- to ensure that network tasks are handled on the network while the CPU does what it does best: maintain the virtual environment.

But not everyone moves that capability directly onto the network. A company called Netronome has devised what it calls the "network flow processor" (NFP), a PCIe-based add-in card that handles everything from I/O virtualization to packet inspection and security. The idea is to provide a level of intelligent data handling that can be implemented across servers, appliances and other network elements.

"The flow processor provides a high-speed network interface on one side, then a virtualized IOV interface to the system, with programmable DPI, packet capture, load-balancing and zero-copy drivers to the multicore CPUs," says Jarrod Siket, senior vice president of sales and marketing at Netronome. "These capabilities ensure that network I/O is no longer the bottleneck of virtualized servers and appliances, allowing the multicore CPUs to realize maximum performance and application processing by offloading burdensome networking tasks."

Of course, infrastructure does not stop at the data center wall, a particular truism as virtualization gives way to the cloud. Shoring up wide-area capabilities is starting to take center stage as users come to expect rapid access to their data no matter where it is housed.

Nowhere is this need more conspicuous than at public cloud providers, which stake their reputations on their ability to mirror traditional data center service at lower cost. NewServers Inc., for example, recently bolstered its capabilities with accelerators from Silver Peak. Still, performance gains are spotty, according to founder J.P. Gagne.

"The performance benefit varies from application to application," he says. "There are many accelerator models ranging in cost from $3,000 to $50,000. These have different configurations of processors, RAM and hard drive arrays. The customer must purchase a device and install it in their data center or office, then lease or co-locate a device in our data center. These create an accelerated VPN from LAN to LAN over the WAN. (But) this does cost performance. WAN acceleration does not create actual LAN performance  over a WAN. That's impossible. It works around some of the limitations of TCP to make applications designed for a LAN usable over a WAN."

Probably the most difficult aspect of preparing enterprise infrastructure for virtualization and the cloud is that there are no proven templates to work with. Infrastructure is rarely the product of design, having evolved over time to meet the immediate needs of the moment rather than some grand design.

That means no single solution will provide the magic bullet to suddenly propel you to dynamic, infinitely scalable resources.

But even if the answers prove elusive, the rewards for getting it right will be astounding.