Zac -
07 Sep 2008
Your chapter on output buffering is one of the best pieces I've read on the topic. However, I would like to know how you are measuing _performance_ since you say there is a "performance improvement" when saving up the data and kicking it out all at once.
I agree (because of the subtlties of ethernet frame, IP packet and TCP packet fragmentation and reconstruction), that network utilization is better (fewer packets for the same amount of data which decreases network use).
However, I decided to set up a small benchmarking script to test if buffering really made the page faster or more scalable (the two forms of _performance_ that is most critical to high-profiles sites).
On my system (dual PIII running FreeBSD/Apache), I found that output buffering slowed down the same code by 400%. Here's my script:
<?php /** Output Buffering Benchmark Test **/
$started = microtime(true);
if($_REQUEST['doit'] == 'ob') {
ob_start();
print_stuff();
ob_end_flush();
}
else if($_REQUEST['doit'] == 'norm') {
print_stuff();
}
else if($_REQUEST['doit'] == 'string') {
echo get_string();
}
else {
?><p><a href="test.php?doit=ob">Run OB Test</a></p>
<p><a href="test.php?doit=norm">Run Non-OB Test</a></p>
<p><a href="test.php?doit=string">Run String Buffer Test</a></p><?php
}
echo 'Run Time: '.number_format((microtime(true)-$started),8).' seconds.';
function print_stuff() {
echo '<pre>';
for($i = 0; $i < 600; $i++) {
for($j = 0; $j < 60; $j++) { echo '.'; }
echo "\n";
}
echo '</pre>';
}
function get_string() {
$buffer = '<pre>';
for($i = 0; $i < 600; $i++) {
for($j = 0; $j < 60; $j++) { $buffer .= '.'; }
$buffer .= "\n";
}
return($buffer.'</pre>');
}
?>
The results of my tests (averages over 10+ executions):
No buffering: 0.084 seconds
Output Buffering: 0.327 seconds
String Buffering: 0.330 seconds
What I found was actually more dramatic than I expected. I guessed that OB would be a little slower--maybe double in a really bad situation. But, I didn't guess that it would be nearly 4 times slower.
The last tes
