> The best and the only scalable approach is to use recent Kernel 4.5+ and use level-triggered events with EPOLLEXCLUSIVE flag. This will ensure only one thread

> The best and the only scalable approach is to use recent Kernel 4.5+ and use level-triggered events with EPOLLEXCLUSIVE flag. This will ensure only one thread is woken for an event, avoid "thundering herd" issue and scale properly across multiple CPU's Exactly use EPOLLEXCLUSIVE for accept, that seems to work and has been in the kernel for more than a year. For reads, how reasonable is to even share the file descriptor with multiple threads. Just hand it to a thread and let that thread (possibly bound to a CPU core for possible more performance boost) handle it from then on. Am I crazy to be surprised by that architecture choice?