A review of the key properties of network virtualization can inform your planning and help in requirements generation as you architect new systems. The best source of information I’ve found on network virtualization is at Nicira, a firm anyone with an infrastructure should be paying attention to now.

The following is drawn from their paper on The Seven Properties of Network Virtualization”

1. Independence from network hardware
In the emerging multi-tenant cloud, the old rules of vendor lock-in are rapidly changing. A network virtualization platform must be able to operate on top of any network hardware, much like x86 server hypervisors work on top of any server. This independence means the physical network can be supplied by any combination of hardware vendors. Over time, newer architectures that better support virtualization as well as commodity options are becoming available, further improving the capital efficiency of cloud.

2. Faithful reproduction of the physical network service model
The vast bulk of enterprise applications have not been written as web applications, and the cost/payback ratio of rewriting tens of billions of dollars of application development is neither realistic nor even possible. Therefore, a network virtualization platform must be able to support any workload that runs within a physical environment today. In order to do so, it must recreate Layer 2 and Layer 3 semantics fully, including support for broadcast and multicast. In addition it must be able to offer higher-level in-network services that are used in networks today such as ACLs, load balancing, and WAN optimization.

It is also important that the virtual network solution fully virtualize the network address space. Commonly, virtual networks are migrated from or integrated with physical environments where it is not possible to change the current addresses of the VMs. Therefore, it is important that a virtual network environment not dictate or limit the addresses that can be used within the virtual networks, and that it allows overlapping IP and MAC addresses between virtual networks.

3. Follow operational model of compute virtualization
A key property of compute virtualization is the ability to treat a VM as soft state, meaning it can be moved, paused, resumed, snapshotted, and rewound to a previous configuration. In order to integrate seamlessly in a virtualized environment, a network virtualization solution must support the same control and flexibility for virtual networks.

4. Compatible with any hypervisor platform
Network virtualization platforms must also be able to work with the full range of server hypervisors, including Xen, XenServer, KVM, ESX, and HyperV, providing the ability to control virtualized network connectivity across any network substrate as well as between hypervisor environments. This “any-to-any” paradigm shift provides for:

• Ÿ More effective utilization of existing network investments,
• Ÿ Cost and management reduction of new, Layer 3 fabric innovations,
• Ÿ Workload portability from enterprise to cloud service provider environments.

5. Secure isolation between virtual networks, the physical network, and the control plane
The promise of multi-tenancy requires maximum utilization of compute, storage and network assets through sharing of the physical infrastructure. It is important that a network virtualization platform maintain this consolidation while still providing the isolation needed by regulatory compliance standards such as PCI or FINRA, as well as provide the same security guarantees of compute virtualization.Like compute virtualization, a network virtualization platform should provide strict address isolation between virtual networks (meaning one virtual network cannot inadvertently address another) as well address isolation between the virtual networks and the physical network. This last property removes the physical network as an attack target unless the virtualization platform itself is undermined.

6. Cloud performance and scale
Cloud drives a significant increase in the scale of tenants, servers, and applications supported in a single data center. However, current networks are still bound by the physical limitations of networks, especially VLANs (which are limited to 4,096). VLANS were designed during an earlier era before server virtualization dramatically increased the requirements for the numbers of virtually isolated environments. Network virtualization must support considerably larger scale deployments with tens thousands, or even hundreds of thousands of virtual networks. This not only enables a larger number of tenants, but also support critical services like disaster recovery, data center utilization, etc., which outstrip current limitations.

A virtual network solution should also not introduce any chokepoints or single points of failure into the network. This roughly entails that to all components for the solution must be fully distributed, and all network paths should support multi-pathing and failover. Finally, a network virtualization solution should also not significantly impact data path performance. The number of lookups on the data path required to implemented network virtualization is similar to what data paths perform today. It is possible to implement full network virtualization in software at the edge of the network and still perform at full 10G line rates.

7. Programmatic network provisioning and control
Traditionally, networks are configured one device at a time, although this can be accelerated through the development of scripts (which emulate individual configuration). Current approaches make network configuration slow, error prone and open to security holes through a mistaken keystroke. In a large-scale cloud environment, this introduces a level of fragility and manual configuration costs that hurt service velocity and/or profitability.

A network virtualization solution should provide full control over all virtual network resources and allow for these resources to be managed programmatically. This allows the provisioning to happen at the service level versus the element level significantly simplifying provisioning logic and any disruption that might occur due to physical network node failure. The programmatic API should provide full access to management and configuration of a virtual network to not only support dynamic provisioning at cloud time scales, but also the ability to introduce and configure services on the fly.

Concluding Thoughts
The seven key features above are a great starting point for requirements for your enterprise architecture. The good news is that you can enjoy all these features of network virtualization without significant change. The only thing it really requires is an understanding of this new approach and access to the technical thought leadership.

For more on this topic a great place to start your research is with Nicira.

This post by BobGourley was first published at CTOvision.com.