Support for OpenStack Kilo and Latest Open vSwitch Boost VM Performance in New Open Network Platform v1.4

By Dana Nehama, Sr. Product Marketing Manager, Network Platforms Group (NPG), Intel

I’m very pleased to announce the availability of a new release of the Intel® Open Network Platform – our integrated, open source NFV and SDN infrastructure reference architecture.

The latest version (v1.4) now offers improved virtual machine platform and communications performance through the integration of OpenStack Kilo, DPDK, and Open vSwitch among other software ingredients.

ONP is the Intel® open source SDN/NFV reference software platform that integrates the main software components necessary for NFV so that Intel Network Builders ecosystem - or any NFV/SDN software developer - have access to a reference high-performance NFV infrastructure optimized for Intel Architecture.

One of the goals of ONP is to improve the deployability of the software components (such as OpenStack or OpenDaylight) by integrating recent releases and, in that process, addressing feature gaps, fixing bugs, testing the software and contributing development work back to the open source community.

The other major goal is to deliver the highest performance software possible, and with v1.4 there is a significant performance improvement for VMs thanks to new features in OpenStack Kilo and Open vSwitch 2.3.90.

Kilo Brings Enhanced Platform Awareness

Advancements in Enhanced Platform Awareness (EPA) in OpenStack Kilo will have a significant impact on the scalability of NFV solutions, predictability and improved performance of virtual machines. EPA is composed of several technologies that expose hardware attributes to the NFV orchestration software to improve performance. For example, with CPU pinning a VM process can be “pinned” to a particular CPU core.

The Non-Uniform Memory Architecture (NUMA) topology filter is a complementary capability that enables memory resources proximity to the CPU core resulting with lower jitter and latency. I/O aware NUMA scheduling adds the ability to select the optimal socket based on the I/O device requirements. Now, with all those capabilities in place, a VNF can pin a high-performance process to a core and ensure that it is connected locally to the relevant I/O, and has priority access to the highest performing memory. All of this leads to more predictable workload performance and improves providers’ ability to meet SLA.

If you’d like more information on EPA, my colleague Adrian Hoban has written a blog post and whitepaper that offers a great exploration of the topic.

Improved Virtual Switching

Another enhancement is the integration of Open vSwitch 2.3.90 with Data Plane Development Kit (DPDK) release 1.8 libraries. The addition of DPDK’s packet acceleration technology to the virtual switch through user space improves VM-to-VM data performance. ONP Release 1.4 also adds support for VFIO which secures the user space driver environment more effectively.

I would like to take the opportunity and mention the significant milestone of OPNFV Arno release, which represents an important advancement towards NFV acceleration and adoption. Intel ONP contributes innovation with partners into OPNFV and in parallel will continuously consume technologies delivered by OPNFV. For additional information on Arno go to https://networkbuilders.intel.com/onp.

I am proud to announce Intel ONP was awarded Most Innovative NFV Product Strategy (Vendor) by Light Reading Leading Lights Awards 2015 on June 8, 2015. The award is given to the technology vendor that has devised the most innovative network functions virtualization (NFV) product strategy during the past year. For more information, please view the announcement here.

ONP 1.4 delivers innovation and significant new value to NFV software developers and I encourage all of you to check out this new reference release. The first step is to download our new v1.4 data sheet.