I’ve been quiet for a while as I’ve been off working on different things. Recently I started digging into the area of Network Functions Virtualization (NFV).
As most of us likely do, when I need to learn new things, the 1st thing I do is go search the Internet for information on the topic. My search on NFV led me to a growing number of papers on using SR-IOV for NFV.
This made sense to me, given that SR-IOV bypasses the hypervisor and virtual switch and can provide better performance by doing so. Yet when I read these documents the numbers published for performance all looked great, however many of them were for just one or two VM/VNF – this didn’t seem like a very valid NFV use case.
When I think of NFV I think of several Virtual Network Functions (VNF) VM’s (or perhaps containers) running all on the same platform, with traffic sometimes going from one VNF to another in a service chain.
Thus began a 6 month experiment in which Brian Johnson and I setup test environments to see how performance scaled when using SR-IOV compared to using Open vSwitch with DPDK enhancements. We have created a technical paper on our results that we hope can provide guidance when looking at Ethernet solutions for your NFV needs.
I hope you find it of use. If you do please comment so we know if folks actually read these docs
The paper is available here: http://www.intel.com/content/dam/www/public/us/en/documents/technology-briefs/sr-iov-nfv-tech-brief.pdf