The networking and telecom industries are undergoing a transformation due to modern advances that promise cost savings and flexibility. Network Functions Virtualization (NFV) architecture enables open platform ecosystems to replace dedicated hardware with commodity servers and virtualization.
Network Functions Virtualization offers a key solution by delivering an alternative platform to create services and applications on flexible network architectures. However, if the proper software architecture is not deployed on servers with the transition to new generation architecture, competitive advantage and time-to-market is lost because performance will decrease versus legacy equipment. If you are developing a Linux-based server networking application for the new generation of your architecture, beware of storms ahead.
A requirement of this software revolution is that the same or greater performance be achieved with virtualization as with legacy physical equipment to realize cost savings and flexibility, but this is not natively possible. Hypervisors, virtual switches and virtual machines that build the cloud application foundation add layers of software overhead and associated bottlenecks.
A picture is worth a thousand words, so let’s see some examples. When designing physical appliances, here are some common NFV performance bottlenecks.
When migrating to virtualization, the bottlenecks can multiply. And NFV performance bottlenecks can further expand with VM to VM communications, and more.
Packet processing software becomes increasingly relevant amid this software transition with technology designed to bypass NFV performance bottlenecks before they occur. When evaluating packet processing software, it is important to check for the following to enable a cost effective value proposition, and smooth transition to future requirements:
- Equal physical and virtual performance
- Transparent, no change necessary to OS, hypervisor, virtual switch and management
- Available across multi-vendor processors, NICs and hardware platforms
If the above considerations are accounted for, deployment of packet processing software can be easy and universal with drop in portability to any version of Linux or hypervisor, without changes to existing environments. To give an example, by accelerating the virtual switch, such as Open vSwitch, performance can increase by over 10X.
When designed in early, packet processing software can resolve NFV performance bottlenecks before they occur to enable the true promise of NFV. It can be likened to a turbo boost for Linux networking: OEM fitment with aftermarket performance.