More on windows performance

Hi,

I've compiled the latest php out of CVS and I'm having the same (wacky)
performance trouble that I've had before. I have cygwin on my machine
that I'm currently using 'time' to test php.exe with my ~100k files of
xes.

The Release_TS_inline numbers appear to be only slightly faster than the
Release_TS numbers. The binary windows php-4.0.0 release is faster
still.

Using the command:
time ./php.exe speed-100k-html.php > dump

For my file with just 100,000 xes, I get this time:
0m7.750s real
0m0.031s user
0m0.031s sys

For my file that <? print ?>s 100,000 chars, I get this time:
0m3.016s real
0m0.046s user
0m0.015s sys

And for php3 I get:
0m0.250s real
0m0.031s user
0m0.031s sys


Perhaps one of you can decipher these numbers... It seems to me that
the user+sys time only adds up to over 0.06 seconds. Which I believe
would be the time that something like this SHOULD actually take.
However, the actual elapsed time is odd because: a) its so slow, b) the
print statements are so much faster than raw characters.

If it would help for me to mail anyone the files I am testing, I will...
I would attatch them here, but they are 200k and it is so easy to
recreate such things in vi ('100ixESCyy999pp').

When I compiled from cvs, the only change I made was adding
ext/standard/scanf.[ch] to the project, because a couple files
referenced php_sscanf_internal.

If you can lend any suggestions for what I might do to get this time
down, please, mail 'em off to me.

-joel

> Perhaps one of you can decipher these numbers... It seems to me that
> the user+sys time only adds up to over 0.06 seconds. Which I believe
> would be the time that something like this SHOULD actually take.
> However, the actual elapsed time is odd because: a) its so slow, b) the
> print statements are so much faster than raw characters.

I suspect the speed is due to the server not having to parse through all
those characters. If you have any type of parsing happening on your raw
HTML files -- for SSI, etc. -- the server has to parse through every single
character looking for control codes. With the single print statement in a
loop, it is able to simply execute the command.

I have no figures to back up this supposition, but it has a "common sense"
logic to it. See how long it takes (you, not a computer) to find the string
"<?" in a file of 100,000 random characters, then see how long it takes to
find the same string in the file:
<? while ($i < 100000):
echo("*");
$i++;
endwhile ?>
It parses much faster, then just spits out the characters. Has anyone done
a test of raw HTML output speed with all parsing, including SSI, disabled?

Aye. I will describe both files again:

1) The 'html' file looks like this:
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
..... etc
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

2) The 'print' file looks like this:
<?
print 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx';
print 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx';
..... etc
print 'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx';
?>

(Both actually contain 1000 lines of 100 xes.)

File 2 is twice as fast as file 1. Both are no where near as fast as linux.
I have repeated similar observations on 3 other computers. Windows NT 4sp4,
4sp6, 2000. Yet I'm not sure anyone has been able to successfully repeat my
experiment.

Here's some comparisons:
$ time cat speed-100k-html.php > dumpfile
real 0m0.094s
$ time ./php3.exe speed-100k-html.php > dumpfile
real 0m0.235s
$ time ./php4.exe speed-100k-html.php > dumpfile
real 0m6.625s

-joel


Drew Kime wrote:

> > Perhaps one of you can decipher these numbers... It seems to me that
> > the user+sys time only adds up to over 0.06 seconds. Which I believe
> > would be the time that something like this SHOULD actually take.
> > However, the actual elapsed time is odd because: a) its so slow, b) the
> > print statements are so much faster than raw characters.
>
> I suspect the speed is due to the server not having to parse through all
> those characters. If you have any type of parsing happening on your raw
> HTML files -- for SSI, etc. -- the server has to parse through every single
> character looking for control codes. With the single print statement in a
> loop, it is able to simply execute the command.
>
> I have no figures to back up this supposition, but it has a "common sense"
> logic to it. See how long it takes (you, not a computer) to find the string
> "<?" in a file of 100,000 random characters, then see how long it takes to
> find the same string in the file:
> <? while ($i < 100000):
> echo("*");
> $i++;
> endwhile ?>
> It parses much faster, then just spits out the characters. Has anyone done
> a test of raw HTML output speed with all parsing, including SSI, disabled?
>
> --
> PHP Windows Mailing List (http://www.php.net/)
> To unsubscribe, e-mail: php-windows-unsubscribe@lists.php.net
> For additional commands, e-mail: php-windows-help@lists.php.net
> To contact the list administrators, e-mail: php-list-admin@lists.php.net