The market is still early, marking a critical evolutionary step for the communications infrastructure that is relied on by almost everyone and everything. There are a number of vendors, new and old, who are each working to deliver what they see as the optimal network virtualization (NV) offerings. There are some that believe changes should be driven by the network, itself, while others feel the applications should be the main influencing factor. It is too early to tell which approach will win, which will lose or which will morph into something yet to be defined.
The approaches that vendors are taking are heavily dependent on where the NV solutions are going to be deployed. As noted earlier, many organizations are deploying NV solutions within their Data Center environments, followed by private, hybrid and public cloud environments. In addition, there are vendors specializing in offering solutions for infrastructure as a service (IaaS) or WAN deployments.
Network Virtualization Architectures in the Data Center
There are really two main approaches to data center deployments:
- Directly programming the fabric. This approach takes explicit control over the switches, creating and coordinating the virtual networks by directly programming the virtual switch and physical switches. Typically this approach requires customers to upgrade all their physical switches to support the appropriate protocols; a flow-control protocol, such as OpenFlow, or a vendor’s proprietary protocol, will be used to manage the network.
- Developing a network overlay. There are a variety of ways in which an overlay network is accomplished; the most common is to modify or replace the virtual switch. In some cases, kernel modules are inserted into the hypervisor (most commonly with KVMas the hypervisor). In situations where the virtual switch cannot be replaced and the hypervisor is proprietary (VMware ESXi), vendors may choose to use a VM running a virtual switch instance as the terminating point for virtual networks.