OPNFV Summit 2015

Many OPNFV developers building Openstack clouds at scale have a “love-hate” relationship with OVS. They love the flexibility and elasticity offered by a distributed virtual switch operating within each server, but hate the reality of first-gen OVS implementations well-known to be a bottleneck for cloud network performance and scalability. As performance-sensitive VNFs keep pushing for higher data forwarding performance across the NFV network infrastructure, it becomes critical to improve OVS performance without compromising flexibility, network programmability, and cost.

In this session, Mellanox will present a novel way to offload the entire OVS dataplane onto the embedded switch in the server NIC. This approach can not only boost server I/O performance to near line-rate, be it 10G, 40G, or 100G, but also doing so at a fraction of the CPU load needed by existing OVS implementation.