Performance of select/poll vs asynchronous I/O

Question

From a performance standpoint, which one is better? select/poll or asynchronous I/O? My earlier impression was select/poll repeatedly asks the kernel for data, whereas asynchronous I/O relies on kernel's notification for data availability. However, I have noticed that select/poll also relies on kernel notifications. So, I believe from a performance standpoint both are same. The only difference is that select/poll blocks whereas asynchronous I/O does not. Am I correct or am I missing something?

Answer 1

With async I/O, you have to loop continiously and check to see if there is new data to read periodically. This makes it CPU intensive. Select/poll simply blocks, taking up no CPU power. It does not loop internally.

Answer 2

select/poll also relies on kernel notification for ready filedeskriptors. But the disadvantage of select/poll is that they block as soon they are called because the Systemcall-Handler runs in the Kernel-Space.

Real asynchronous I/O is achieved via LibAIO (on Linux) and IOCP on Windows. As far as i know they dont block the calling process/thread in der User Space and they allow real overlapped I/O.

That means asynchronous Non Blocking I/O (LibAIO & IOCP) is faster, because it does not block the calling Thread and they allow real overlapped I/O. Select/poll are also asynchronous, but they are Asynchronous Blocking. And btw select and poll suffer from other specific problems so that they cant scale that well.

Hope i could help u. (I am a newbie on this too :))