RE: Patch to eliminate user space memory copying in 'cp'

[Spinka, Kristofer]

  I agree that there wasn't any notable "copy" I/O performance increase,
but the CPU cycles and memory I/O are significantly reduced.

  So while most modern machines' have CPUs and memory buses that have
plenty of room for inefficient user-space copies on a quiescent machine,
it's the efficiency I was looking for.  I think on a heavily loaded
machine or with a serious disk subsystem the difference would be more
apparent to the user.

   /kristofer

> -----Original Message-----
> From: Jim Meyering [mailto:address@hidden
> Sent: Saturday, April 05, 2003 1:29 PM
> To: Spinka, Kristofer
> Cc: address@hidden
> Subject: Re: Patch to eliminate user space memory copying in 'cp'
> 
> Jim Meyering <address@hidden> wrote:
> > "Spinka, Kristofer" <address@hidden> wrote:
> >>   You will have to incorporate the check for sendfile, but
otherwise
> >> this patch should avoid the most common kernel<->user space
read/write
> >> transitions.
> >
> > Thank you for that patch.  It looks like a fine optimization.
> >
> > I'll work on integrating it after coreutils-5.0.
> 
> After a few simple tests, I don't see any performance
> benefit to using sendfile.  Do you have evidence that
> using sendfile is worthwhile?