Re: [boost] Non-blocking I/O

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2001-03-04 12:00:44

I've gotten good speedups using non-blocking I/O on NT, but only in
the case where I could overlap expensive computations with the I/O.

From: Beman Dawes <bdawes_at_[hidden]>
> At 03:13 PM 3/3/2001 +0000, pinkfloydhomer_at_[hidden] wrote:
> 
>  >Would a non-blocking I/O library addition be off topic in Boost, or
>  >wouldn't it?
> 
> It certainly wouldn't be off topic for Boost.
> 
> But I wonder if it would actually result in a performance improvement on 
> most modern operating systems.
> 
> Don't most OS's already do anticipatory buffering?  Have you run timings?
> 
> I ran some timings a few years ago on Win NT, and found that the compiler's 
> runtime library was actually slightly more efficient than explicitly 
> calling the native Win NT non-blocking I/O.  The build-in OS I/O buffering 
> was more efficient that anything I could do in user code, perhaps because 
> it remaps memory and does other supervisor mode tricks you can't do in user 
> mode.
> 
> But if your timings show significant performance enhancements, non-blocking 
> I/O might well be of interest.
> 
> --Beman
> 
> 
>  
> 
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/ 
> 
> 
>