io_uring: path to zerocopy I/O

With I/O devices getting faster each year, memory copies are becoming more and more expensive, wasting a lot of CPU cycles and being a burden to the memory subsystem. The problem goes even deeper, as device-to-device transfers usually require intermediate steps, pumping the data through the RAM. We’re going to explore what io_uring has to offer and plans in the areas of zerocopy and peer-to-peer I/O.

Pavel Begunkov