Tariq Toukan will discuss creating multiple PFs for a single NIC in a multi-socket server, each with its own PCIE interface.

Why do we need this you ask? Imagine a NUMA machine... you plug in a NIC.... as it sits today, the NIC gets connected on one of the NUMA nodes. If your application is running/scheduled on a different numa node then network traffic traverses over the internal bus between the sockets. The consequences are observed as significant latency, CPU utilization, and decreased network throughput.

In this talk Tariq describes the Multi-PF technology under the Nvidia netdev driver which overcomes these challenges. The Multi-PF approach enables several CPUs within a multi-socket server to connect directly to the network, each through its own dedicated PCIe interface.

For more information: netdevconf.info/0x18/24

cheers, jamal