1f604 / liburing_b3sum Goto Github PK

Currently the consumer thread spins on an atomic (and likewise the producer thread is also effectively a busy wait).

This may be wasteful of CPU. We could perhaps reduce CPU usage by switching to cv.wait.

This may require splitting up the producer thread into 2 new threads: one that issues requests in a loop and one that checks for completions in a loop. We need to investigate whether this is okay to do (i.e. does io_uring allow it?).

If there is enough demand for this then I can implement it, but currently since nobody has complained about it I'm going to assume that it's not a problem.

Currently, the program just hashes a single file.

However, we could make the program hash multiple files.

This could potentially bring some performance benefits since we can avoid the overhead of setting up the io_uring and the ring buffer for each file, and possibly open the next file before we have finished processing the current file, and send the request for the next file before we're done reading the current file - I'm pretty sure this will result in some amount of performance gain, but I don't know how much - need to benchmark it.

If there is enough demand for this then I will implement it. Since nobody has complained about this I'm going to assume that it's a non-issue.