I believe that the full value of Network Functions Virtualization (NFV) is best realized using Pure-Play Virtualization. We define pure-play virtualization as software Virtual Network Functions (VNFs) running in a standard virtualization environment (e.g. KVM and OpenStack) hosted on open servers. Pure-play virtualization lets service providers achieve the benefits of rapid composition and activation of innovative services that can be hosted on the right-sized platform at the appropriate location in the network (edge, CO or datacenter).
I read recently a post where a hardware supplier stated that Pure-Play is impossible due to high cost and low performance. They and other proponents of this position state that pure software running on open hardware will either be too expensive, fail to meet throughput and latency requirements, or both. They say the only way to address virtualization at the edge of the network is with a hybrid approach: either a traditional NID with an added compute element, or most recently with an SFP with an internal compute element.
You’d be silly to take the bet because we have already proven that pure-play solutions can meet cost and performance requirements – in the lab and in the field.
Back in May we published test results that showed gigabit throughput for traffic for Carrier Ethernet and routing VNFs running on a standard low-end server using an Intel Atom processor.
In addition, in July Masergy announced its Virtual f(n) service, which is based on pure-play software from Overture, Brocade and Fortinet. Masergy CTO Tim Naramore emphasized the benefits of an open ecosystem of software components running on a standard server – not a hybrid approach with specialized forwarding elements. Using a standard server means that Masergy can rightsize the platform for the application and deployment location now and in the future.
The alternative to pure-play is a hybrid approach that combines dedicated forwarding elements with a limited compute platform. This hybrid approach is a step in the right direction in that it provides for hosting of software VNFs. But it is a bigger step in the wrong direction because the compute platform is locked into to the hardware, which limits service provider choice. It also ties the layer 2 Carrier Ethernet functionality to that specific piece of hardware, limiting future growth in features, as well as preventing deployment in fully virtualized environments i.e. the cloud.
A hybrid approach limits the ability to scale and update and it dictates where each function must be deployed.
Why limit your network and your business? Choose a pure-play virtualization approach that can meet the cost and performance requirements today, while providing an open platform for future growth.