There has been a lot of recent interest in industry efforts towards virtualizing the network to match the transformational results realized by virtualized computing. The argument goes, “we now have all this great virtualized compute capacity that give us the ability to better match services with demand, but the network remains stuck in the past as the LAN/WAN services are still deployed and provisioned in more static ways.” There is some truth to this argument along with some promising new directions (like OpenFlow), but as far as making existing the networks of applications of today and the foreseeable future run better, private clouds and technologies like VDI can be flexibly deployed and optimized using technology available now.
A key consideration, and an essential component of any enterprise cloud that hosts critical services, is how to intercept traffic in the data center so that the right protocol traffic can be optimized for each group of applications while still maintaining enough “on-demand” capability so that your cloud can handle traffic growth, new architectures (Citrix ICA optimization, for example) and avoid any brittleness that might go along with static provisioning. One approach is VRF-aware WCCP: a way to hand-off traffic to optimizers as it travels to these independently provisioned services at lower layers in enterprise network. For example, IT for different business units might be provisioned as separate collections of VMs on separate VLANs and subnets, each with its own IPv4/IPv6 routing table. This approach allows out-of-path interception at core/aggregation layer where traffic can be better abstracted.
If you are involved in planning or thinking about your next private cloud deployment, a straightforward discussion of how to actually make these decisions is discussed in Chapter 6 of a new book by our own Dr. Steve Smoot and Nam Kee Tan. The book, Private Cloud Computing: Consolidation, Virtualization, and Service-Oriented Infrastructure, recently reached the top 50 in its category on Amazon and covers a whole lot more on this topic, including other key considerations like storage architecture, Infrastructure-as-a-Service, and example case studies. There is just the right amount of depth (such as configuration examples) for designing new networking services for clouds -- services like subtenant provisioning that are central to the concept.
In short, optimization can be easily deployed ways that are fully compatible with deploying applications virtually in the enterprise using existing capabilities on your networking gear. Future technologies changing how we think of networking VMs, such as with a FlowVisor, may also enable new ways of optimizing traffic wherever it needs to be. Multi-tenant optimization is real today in a practical sense (it's technically very doable) and it supports the strategic gains of deploying a private cloud: more effectively matching services with demand.