What is difference between PF_RING technology and PACKET_RX_RING

Question

I stumbled upon PF_RING while reading about PACKET_MMAP kernel documentation (https://www.kernel.org/doc/Documentation/networking/packet_mmap.txt)

Can someone explain the difference between the actual technology (implementation details and differences) between PF_RING and PACKET_RX_RING/PACKET_TX_RING in PACKET_MMAP

Answer 1

PF_RING has two very different modes of operation.

• The one called "vanilla" operates "above" driver level, so it should be mostly similar to PACKET_MMAP. They both simply share a buffer between the user application and the network stack. I think PF_RING also discards the packets, so it could be say it's exclusive. PACKET_MMAP, on the contrary, lets the kernel stack process the packets after the copy to userspace.
• The "DNA" or "zero-copy" mode implements kernel bypassing. Instead of copying data to a shared ring buffer, the driver's buffers themselves are shared. This, obviously, requires custom drivers and means no other processes will be able to receive traffic from the affected interfaces. Many commonplace cards are supported. Due to this reduced copying and context switches and interrupts (you can do polling if you want to) you can squeeze quite a lot more of performance. The upstream technology that comes the closest is AF_XDP.

I may have gotten some things wrong (I just Googled for a bit out of curiosity and am by no means an expert in PF_RING), so watch out for other answers. I do think most of what I wrote is accurate.