It's clear that we're in the 'Virtualization Age' in the technology markets, and while there is certainly no shortage of hype in the space, there is significant value in virtualization. Networks are involved in most of the efforts so far -- from server virtualization to cloud computing. But assigning a supporting role to the network can disguise the value of virtualization as applied in the network itself.
All forms of virtualization are based on the assumption that hardware and operations costs are incurred per device, but devices often have the internal resources to support multiple independent missions. If virtualization capabilities are exercised, some operations and hardware costs can be saved. All router virtualization strategies create multiple virtual or logical routers from a single physical device. These virtual router instances are linked in their own networks, creating multiple virtual networks so traffic in one is a 'ship in the night' with respect to traffic in others -- at least in theory. Clearly as router virtualization expands, it increasingly becomes network virtualization.
Virtual device management issues
From the beginning, most routers have supported a primitive form of virtualization in the ability to partition their routing tables. Virtual Routing and Forwarding (VRF) supports multiple instances of routing tables and so effectively creates virtual routers. Another form of router virtualization can be created using MPLS label-switched paths and independent edge routers or 'Label Edge Routers' (LERs), which are at the edge of an MPLS network and whose nodes are Label Switch Routers (LSRs). This effectively create a set of independent router networks that share core network resources. Both of these capabilities are in use today, but they have some common limitations.
One limitation of 'first-generation' router virtualization is that there is no hard partitioning of resources between the virtual instances. Control plane processing requirements are normally increased in proportion to the number of virtual router instances in VRF, and all such processes compete for the processor/memory resources of the same router control plane blade. There may be other interdependencies in handling parallel virtual router instances, depending on the implementation. With segmented-core virtualization using MPLS, you must either deploy independent LERs for each virtual network or design the LER routing tables carefully to prevent cross-routing traffic. These systems may also share trunk facilities among virtual router instances, creating further interdependence of traffic and performance.
Resource interdependency also tends to set an upper limit on the number of virtual router instances and virtual networks that can be deployed. Again, this limit is vendor-specific, but it relates to the fact that virtual router control plane handling requires one type of router blade, and data port/trunk connection requires another. Since there is a fixed amount of space in any rack, the need for more control blades means less space for data blades.
A final issue is the management of virtual devices versus real devices. Operations costs of network equipment are actually 55% of total cost of ownership (TCO), and capital cost is only 45%. If virtual router instances 'appear' as real routers in an operations sense (meaning that they require the same operations processes as a real device), then virtualization will not reduce operations costs. In fact, if managing the virtualization process is anything beyond trivial (which it normally is), router virtualization may actually have higher operations costs that would need to be offset by additional hardware savings.