Vjekoslav Brajkovic wrote:
posting 10 M jobs to the io_service, and starting N threads at io_service::run after that. Timings below are measured on an 8-core machine, I would expect
I suppose you are the one instantiating the connection.
This is apart from any connection, it's measuring io_service as a pure job dispatcher.
I had the same exact symptoms a week ago... not sure if this is the same problem. However instead of using threadpool, I was using thread_group(). The pseudo code looked something like this:
Interesting -- what would be the expected performance difference between a thread group and a pool of threads?
I haven't had time to investigate what was causing the issue.
Do you know if all of io_service's jobs are dispatched through a platform-specific dispatcher, e.g., epoll? Thanks, Rutger