On Wed, Aug 12, 2020 at 10:29 AM David Howells <dhowells@xxxxxxxxxx> wrote: > > Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > Worried about performance? Io-uring will allow you to do all those > > five syscalls (or many more) with just one I/O submission. > > io_uring isn't going to help here. We're talking about synchronous reads. > AIUI, you're adding a couple more syscalls to the list and running stuff in a > side thread to save the effort of going in and out of the kernel five times. > But you still have to pay the set up/tear down costs on the fds and do the > pathwalks. io_uring doesn't magically make that cost disappear. > > io_uring also requires resources such as a kernel accessible ring buffer to > make it work. > > You're proposing making everything else more messy just to avoid a dedicated > syscall. Could you please set out your reasoning for that? a) A dedicated syscall with a complex binary API is a non-trivial maintenance burden. b) The awarded performance boost is not warranted for the use cases it is designed for. Thanks, Miklos
