How Red Hat And Intel are Building the NFV Server of the Future

As part of the launch of the Intel® Open Network Platform Server Reference Design, we’ve asked key development partners to blog on their work with Intel on this reference architecture and their views on network transformation.  Today’s blog is by Chris Wright, Technical Director, SDN and NFV, Red Hat.

Network functions virtualization (NFV) is one of the fastest growing and most talked about technologies in the telecom world today because it can squeeze a lot of cost out of carrier networks by replacing expensive, purpose-built appliances with applications running as virtual machines on Intel® architecture servers.

But these appliances did deliver very high performance – that’s one of the reasons they were expensive. For data plane processing applications such as load balancing, firewalls and virtual CPE, that performance is essential.

The challenge for NFV vendors is to deliver that same level of performance, and that’s why we’ve been working with Intel to develop the NFV Reference Architecture 1.0, which we demonstrated last week at the Intel Developer Forum (IDF) 2014.

Latency Is Key Issue for NFV Servers

Many of today’s NFV server solutions aren’t optimized for data plane processing applications, which results in added packet latency because the virtual infrastructure and virtualized data plane cause packets to be unnecessarily routed through the operating system, hypervisor, and processor.

The NFV Reference Architecture tackles this issue with a combination of Red Hat and Intel software. The reference design is based on the Red Hat Enterprise Linux OpenStack® Platform, which controls the overall NFV infrastructure and provides an open interface for a wide variety of network orchestration tools and virtual network functions software.

Added to that is the Intel® Data Plane Development Kit (®Intel DPDK) Open vSwitch. This set of software libraries accelerates virtual switching and boosts the capability of NFV software to route network packets around the Linux operating system kernel, which Intel has reported can improve packet throughput by up to tenfold.

One of the capabilities built into the operating system is a Kernel-based Virtual Machine (KVM) hypervisor that works with the Open vSwitch to provide high-performance virtual network switching and virtual machines that are optimized for network applications.

The hardware components of the reference design include Intel® Xeon® processor-based servers and 10 gigabit Intel® Ethernet technology to deliver faster processing power and state-of-the art network connectivity.

The big achievement of the reference design is giving NFV vendors a very low latency virtual data plane that is equivalent to an appliance-based data plane. The other achievement, which was very important to Red Hat, is that the solution is open.

As a pioneer in open source software, the Red Hat team felt right at home with Intel’s commitment to the open source software movement.  While we’ve based the NFV Reference Architecture 1.0 on Red Hat Enterprise Linux, we’re also planning to upstream it as a Fedora open source Linux implementation for those developers who would like to have the unsupported version of the code.

When we joined with Intel to work on this project, the goal was to help stimulate the NFV market by providing the performance needed for the most demanding applications.  Last week at IDF, visitors to our booth were able to see our outstanding progress for themselves.