Hielke de Vries

Hi Alex, The 99% increase was with an early version of io_uring and buggy version of liburing. Besides, the performance test tool used had some bugs as well: https://github.com/frevib/io_uring-echo-server/pull/2#discussion_r376728709 So...

This echo server also uses IORING_OP_PROVIDE_BUFFERS, which causes a performance drop: https://twitter.com/hielkedv/status/1255492941960949760?s=21 You could try running it without automatic buffer selection, using your own buffer management implementation or no buffer...

The benchmarks provided are without buffer selection: https://github.com/frevib/io_uring-echo-server/blob/master/benchmarks/benchmarks.md

This could very well the branch, it’s indeed without buffer selection but with fast poll. I haven’t had the time to look for the exact commit hash.

Maybe try this one: https://github.com/frevib/io_uring-echo-server/tree/b003989ecb6343b5815999527310251601531acc This commit is right before buffer selection was implemented.

To understand these performance gains that the echo server is claiming, maybe some extra context is needed. * When io_uring just got out in Kernel 5.1, I used a version...

@alexhultman when I have some time left I will try and find the right commit that does not use IORING_OP_PROVIDE_BUFFERS. I’m quite out of the io_uring scene at the moment,...

Formatting my Samsung evo plus 64gb microsd to NTFS solved the problem, @joaopauloss. I still see some drops from 8000KB/s to 2000KB/s but it goes back to 8000KB/s quickly. These...

This project is indeed quite stale. But if there are any bugfixes or improvements I can merge them. The benchmarking results should be reproducible if you use the exact same...

Hi, this is only a bare minimum show case for io_uring. So please don't expect production-grade stability.