here's a missive fired off by the site admin after he "benchmarked" two
scripts, one written in php and one written in perl/cgi
>
> First of all.
>
> Dude. you're out of your mind. Im serious.
>
> The WHOLE point about why PHP is faster than Perl is because the
> interpreter is compiled into Apache.
he's not running a perl interpreter compiled into Apache.
> Also. the fact that your script caused MySQL to use up all of it's
> connections has -nothing- to do with PHP being compiled into Apache.
I find this terribly difficult to believe, but I'm willing to post my
cgi for review both here and in the DBI list
> We simply do -not- have the hardware to run mod_perl. With our level of
> traffic, we would need a load balanced cluster to handle this.
>
> We skewed nothing. We ran the same apache bench command for both
> scripts. Same number of concurrent requests, same number of times. Do not
> confuse ApacheBench with some useless little tool. This is for serious
> server benchmarking. The fact that we're using gemini table types and
> your database tables are indexed just further shows the limitations of
> Perl.
>
> You simply don't get it. PHP, in all of it's forms, blows perl out of the
> water. I've been writing both since early 1993, and in every case, in
> every instance, PHP crushes perl for speed. That's -why- it was created
> (build in interpreter). That's -why- Zend released the PHP4 engine.
> That's -why- we're running Zend Optimizer. If perl was the shit for doing
> CGI, why would anyone even bother creating things like PHP? That's like
> the Chewbacca website. It makes no sense.
>
> mod_perl is a resource pig. I refuse to install something on a server that
> will make life miserable for everyone else. I've seen GHz machines hauled
> off of their foundations because of mod_perl, while the same server
> running PHP code has no problems whatsoever.
I responded with certain information along these lines:
-=-
> > If you use CGI.pm in many of your mod_perl scripts, you may want to preload
> > CGI.pm and its methods at server startup time. To do this,
> > add the following line to httpd.conf:
> >
> > PerlScript /home/httpd/conf/startup.pl
> >
> > Create the file /home/httpd/conf/startup.pl and put in it all the modules
> > you
> > want to load. Include CGI.pm among them and call its
> > compile() method to precompile its autoloaded methods.
> >
> > #!/usr/local/bin/perl
> >
> > use CGI ();
> > CGI->compile(':all');
> >
> > Change the path to the startup script according to your preferences.
>
> if you're gonna benchmark at least do it right.
>
> don't flap statistics at my face when you've got sandbags tied around the
> feet of all my peasants, and shot each one in the foot as well, please.
>
> Also, Yoda's script is not performing (and from what I can see, can not
> perform ) the same query mine was (again skewing the 'benchmark')
>
> searching his script for 'c' does not return even the same list of maps mine
> does. I feel that a certain degree of *accuracy* is also important in a
> benchmark.
>
> I've also gone to the trouble of doing things such as this:
>
> my( $type, $id, $filename, $title, $size, $reviewfile, $rating, $rated,
> $oldtype );
> $sth->bind_columns(\$type, \$id, \$filename, \$title, \$size,
> \$reviewfile, \$rating);
>
> which binds the results of each return into the same variable reference to
> save on memory and processing while looping through the fetch, instead of
> thrashing the symbol table.
>
> and other things like this
>
> # die with status error if necessary if cgi itself got an error
> if (!param() && cgi_error()) {
> print header(-status=>cgi_error());
> goto FINISH;# don't call exit 0; !!! (unless you LIKE killing your perl
> process over and over, ass-hat) :P
> }
>
> and
>
> if ( $search_obj eq '' )
> {
> # skip the database query
> print end_html;
> goto FINISH; # don't call exit 0!
> }
>
> to prevent stupidity like killing the mod_perl process
>
> so,
> A> the 'benchmark' is invalidated by the virtue of not performing the exact
> same query to the MySQL server. (That shoots the benchmark down right there.
> =:P If the query string isn't exactly the same for both tests how can you
> call it a benchmark?)
>
> B> the 'benchmark' is additionally character-assassinated by not running a
> persistent perl process and CGI process (and DBI process) like you DO have a
> persistent php and php-with-mysql process, causing perl and CGI and DBI to
> re-execute and recompile themselves each time. (what kind of results DID you
> expect doing something like this? =:P)
>
> C> The accuracy of the returned result of the query as performed by each
> script is also in question. (look at the result count reported by each script
> as to how many maps it returned from the query, and tell me something's not
> wrong with one of them. =:P)
>
> try again.
-=-
here's the "benchmark" result he returned to me
We benched your script using ApacheBench. We let it run with 100
concurrent connections. A couple of things happened.
1. MySQL died with a "too many connections" error. Our forums, with 150
users on them, can't even do that.
2. The load average on the box jumped to 11. Not 1 or 2. 11.
3. Yoda has written a PHP search engine which already incorporates all
of
the advanced features for NC. His script ran 229 times faster than
yours,
and the load average never moved above 0.5. MySQL was also perfectly
fine, after being benched under the same conditions.
I strongly recommend at this point that you do not use Perl for
Nalicity. I have included the results of our benchmarks (performed by
Chris), so you can see for yourself. This sort of load would be
unacceptible in the BU environment.
Begin ab log:
Yoda's version:
[root@beyondunreal bin]# ./ab -n100 -c10 -k
http://nalicity.beyondunreal.com/testbed/news2.php?executesearch=1&search
by_titles=on&tSearchText=c&sortby=2&sorttype=1
This is ApacheBench, Version 1.3c <$Revision: 1.45 $> apache-1.3
Copyright (c) 1996 Adam Twiss, Zeus Technology Ltd,
http://www.zeustech.net/
Copyright (c) 1998-2000 The Apache Group, http://www.apache.org/
Server Software: Apache/1.3.22
Server Hostname: nalicity.beyondunreal.com
Server Port: 80
Document Path: /testbed/news2.php?executesearch=1
Document Length: 769 bytes
Concurrency Level: 10
Time taken for tests: 0.574 seconds
Complete requests: 100
Failed requests: 0
Keep-Alive requests: 0
Total transferred: 97206 bytes
HTML transferred: 78438 bytes
Requests per second: 174.22
Transfer rate: 169.35 kb/s received
Connnection Times (ms)
min avg max
Connect: 1 5 17
Processing: 20 44 244
Total: 21 49 261
Fuzzbuster's version:
[root@beyondunreal bin]# ./ab -n100 -c10 -k
http://nalicity.beyondunreal.com/cgi-bin/simplesearch.cgi?searchfor=c
This is ApacheBench, Version 1.3c <$Revision: 1.45 $> apache-1.3
Copyright (c) 1996 Adam Twiss, Zeus Technology Ltd,
http://www.zeustech.net/
Copyright (c) 1998-2000 The Apache Group, http://www.apache.org/
Server Software: Apache/1.3.22
Server Hostname: nalicity.beyondunreal.com
Server Port: 80
Document Path: /cgi-bin/simplesearch.cgi?searchfor=c
Document Length: 698276 bytes
Concurrency Level: 10
Time taken for tests: 115.382 seconds
Complete requests: 100
Failed requests: 0
Keep-Alive requests: 0
Total transferred: 71242100 bytes
HTML transferred: 71215208 bytes
Requests per second: 0.87
Transfer rate: 617.45 kb/s received
Connnection Times (ms)
min avg max
Connect: 1 122 3017
Processing: 9771 11087 11502
Total: 9772 11209 14519
I'm willing to post my script here to see if any of you individuals can
tell me what, if anything, I did wrong with MY script that could have
caused MySQL to die with "too many connections" or whether this is a
problem with DBI and DBD::MySQL in its present form.
I await your response. (with heavy sighs and a great deal of frustration)
--
Scott R. Godin | e-mail : [EMAIL PROTECTED]
Laughing Dragon Services | web : http://www.webdragon.net/
