Re: File System Performance

[Andrew Morton <akpm@zip.com.au>]

Linus Torvalds wrote:
> 
> In article <3BF04289.8FC8B7B7@zip.com.au>,
> Andrew Morton  <akpm@zip.com.au> wrote:
> >
> >It's tar.  It cheats.  It somehow detects that the
> >output is /dev/null, and so it doesn't read the input files.
> 
> Probably the kernel.
> 
> If you do a mmap()+write(), the write() to /dev/null won't even read the
> mmap contents, which in turn will cause the pages to never be brought
> in.
> 
> Anything which uses mmap+write will show this.

Actually, tar _is_ doing funnies with /dev/null.  Changelog says:

1995-12-21  François Pinard  <pinard@iro.umontreal.ca>

        * buffer.c: Rename a few err variables to status.
        * extract.c: Rename a few check variables to status.

        Corrections to speed-up the sizeing pass in Amanda:
        * tar.h: Declare dev_null_output.
        * buffer.c (open_archive): Detect when archive is /dev/null.
        (flush_write): Avoid writing to /dev/null.
        * create.c (dump_file): Do not open file if archive is being
        written to /dev/null, nor read file nor restore times.
        Reported by Greg Maples and Tor Lillqvist.

One wonders why.

-