Re: Perl Quiz of the Week #23

On Fri, 03 Sep 2004 09:47:29 -0400, Mark Jason Dominus <[email protected]> wrote:
> [email protected] (Kurt Starsinic):
> >     % time parens 12 > /dev/null
> >     real    0m4.541s
> >     user    0m4.460s
> >     sys     0m0.012s
> >
> >     The "tail" version has to generate all the output, it just doesn't
> > have to display most of it to the s*l*o*w tty.  Interestingly, the
> > "/dev/null" version isn't that much faster, which reinfoces the idea
> > that the slowness isn't in the I/O subsystem per se, but rather in the
> > tty driver specifically.
> 
> I don't think the > /dev/null version is using the TTY driver.  Why
> would it be?

    Ah, I see that I was unclear.  I'm not implying that the /dev/null
version is using the tty driver.  I'm pointing out that the difference
in times between [sending output to a pipe, where the data actually
has to travel end-to-end, but most of it never touches a tty] and
[sending output to /dev/null, where the data gets copied into the
kernel and thrown away (thus circumventing slightly more than half of
the I/O subsystem)] is very small, which suggests that the I/O
subsystem in and of itself isn't a major contributor to overall
runtime.

> I guess that you will not see much decrease in time if you remove the
> printing from your program.

    Exactly:
        % time parens_no_output 12
        real    0m4.252s
        user    0m4.200s
        sys     0m0.012s

    - Kurt