support io_uring for something close to optimal performance

[matu3ba]

Justification: Should be faster than mmap, see for example https://qdrant.tech/articles/io_uring/ or generally the absence of synchronization.

Usage requires to roll your own synchronization, if necessary. One way of usage without any synchronization is raw memory of word size and reseve 2 values indicating sender and receiver, but this bears down and upsampling cost.

[alexandervanrenen]

Using io_uring would be super interesting. The repo is pretty old, but it would be great to try out some of the newer methods (maybe some of the user space file system thingies like FUSE).

Thanks for the update, I think it could be a good time to revist this benchmark. Personally, I do not have time, but we might have a student working on this again :)