Re: Apache::Session and performance question
Perrin Harkins wrote: On Mon, 28 Aug 2000, Chris Brooks wrote: I went back through the documentation on Apache::Session, Apache::Session::DBIStore, and Apache::DBI, and I haven't found a problem in the way we have implemented this. Does anyone else have suggestions, or has anyone else experienced a similar performance hit? You're not doing anything wrong. As I said before, going to a databse on every request is expensive. It will definitely crush your performance when compared to something really lightweight like serving static files with no db access. You could try using one of the other session stores like FileStore to see if it's any faster for you, or you could try running MySQL on the same machine as the web server, but if you have significant traffic you will eventually need a separate db server machine and a cluster of web servers. The only thing I could suggest for improving the performance of your setup is to make sure you have MySQL properly tuned, with an appropriate index on this table. - Perrin I think Perrin has already touched on this and it may be too bovious, but if you are session managing all requests you are wasting alot of time on non-html / dynamic content such as images, style sheets. Greg Cope
Re: Apache::Session and performance question
Hi again to the list, I went back through the documentation on Apache::Session, Apache::Session::DBIStore, and Apache::DBI, and I haven't found a problem in the way we have implemented this. Does anyone else have suggestions, or has anyone else experienced a similar performance hit? Thanks, Chris Date: Tue, 22 Aug 2000 10:08:11 -0400 To: Perrin Harkins [EMAIL PROTECTED] From: Chris Brooks [EMAIL PROTECTED] CC: [EMAIL PROTECTED] Subject: Re: Apache::Session and performance question Message-ID: [EMAIL PROTECTED] Hi, thanks for the reply, Yes, we are calling the module from http.conf like this: PerlModule Apache::DBI PerlRequire /www/perl/scripts/perl-startup.pl PerlHeaderParserHandler Carescout::SessionHandler The referenced startup script contains the line: Apache::DBI-connect_on_init( 'DBI:mysql:cs', 'nobody', '' ); I added $Apache::DBI::DEBUG = 1; to the startup file, and it doesn't seem to be creating a new connection for every request. Thanks, Chris Perrin Harkins wrote: On Mon, 21 Aug 2000, Chris Brooks wrote: We have a fairly simple handler responsible for maintaining state on our web server. Unfortunately, when we activate it, server performance drops to about 1/10th of what it is without. After going through the handler and commenting out parts and benchmarking (rinse, repeat), Apache::Session appears to be the culprit. I can't believe that the use of this module should have so great an effect. We have to be doing something incorrectly. Any advice would be greatly appreciated. You are using Apache::DBI, right? Going to the database on every single request is going to be expensive. If the rest of your application doesn't normally hit the database it makes sense for this to have a significant impact. - Perrin -- Chris Brooks Director of Technology CareScout.com phone: (781) 431-7033 x 342
Re: Apache::Session and performance question
On Mon, 28 Aug 2000, Chris Brooks wrote: I went back through the documentation on Apache::Session, Apache::Session::DBIStore, and Apache::DBI, and I haven't found a problem in the way we have implemented this. Does anyone else have suggestions, or has anyone else experienced a similar performance hit? You're not doing anything wrong. As I said before, going to a databse on every request is expensive. It will definitely crush your performance when compared to something really lightweight like serving static files with no db access. You could try using one of the other session stores like FileStore to see if it's any faster for you, or you could try running MySQL on the same machine as the web server, but if you have significant traffic you will eventually need a separate db server machine and a cluster of web servers. The only thing I could suggest for improving the performance of your setup is to make sure you have MySQL properly tuned, with an appropriate index on this table. - Perrin
Re: Apache::Session and performance question
Perrin, Thanks for the replies. Adding an index made a significant improvement on performance -- it's still three or four times slower than without Apache::Session, but much faster than without the index. Thanks again, Chris Perrin Harkins wrote: On Mon, 28 Aug 2000, Chris Brooks wrote: I went back through the documentation on Apache::Session, Apache::Session::DBIStore, and Apache::DBI, and I haven't found a problem in the way we have implemented this. Does anyone else have suggestions, or has anyone else experienced a similar performance hit? You're not doing anything wrong. As I said before, going to a databse on every request is expensive. It will definitely crush your performance when compared to something really lightweight like serving static files with no db access. You could try using one of the other session stores like FileStore to see if it's any faster for you, or you could try running MySQL on the same machine as the web server, but if you have significant traffic you will eventually need a separate db server machine and a cluster of web servers. The only thing I could suggest for improving the performance of your setup is to make sure you have MySQL properly tuned, with an appropriate index on this table. - Perrin -- Chris Brooks Director of Technology CareScout.com phone: (781) 431-7033 x 342
Apache::Session and performance question
Hi all,
We have a fairly simple handler responsible for maintaining
state on our web server. Unfortunately, when we activate
it, server performance drops to about 1/10th of what it is
without. After going through the handler and commenting
out parts and benchmarking (rinse, repeat), Apache::Session
appears to be the culprit.
I can't believe that the use of this module should have so
great an effect. We have to be doing something
incorrectly. Any advice would be greatly appreciated.
package Carescout::SessionHandler;
use strict;
use Apache;
use Apache::Session::DBI;
sub handler {
my ( $r ) = shift;
my ( $cookie ) = $r-header_in('Cookie') || undef;
$cookie =~ s/SESSION_ID=(\w*)/$1/o;
my ( %session );
tie %session, 'Apache::Session::DBI', $cookie,
{DataSource = 'dbi:mysql:cs',
UserName = 'nobody',
Password = ''
};
if ( ! $session{ acl } ) {
$session{ acl } = '0';
}
if ( ! $session{ uid } ) {
$session{ uid } = '99';
}
if ( ! $session{ services } ) {
$session{ services } = '0';
}
$r-subprocess_env( 'SESSION_ID' = $session{
_session_id } );
$r-subprocess_env( 'UID' = $session{ uid } );
$r-subprocess_env( 'ACL' = $session{ acl } );
$r-subprocess_env( 'SERVICES' = $session{ services }
);
my ( $session_cookie ) = "SESSION_ID=$session{
_session_id };path=/;";
tied( %session)-make_modified();
untie %session;# not sure if we need this
$r-header_out("Set-Cookie" = $session_cookie ) if !
$cookie;
}
1;
__END__
Thanks,
Chris
Re: Apache::Session and performance question
On Mon, 21 Aug 2000, Chris Brooks wrote: We have a fairly simple handler responsible for maintaining state on our web server. Unfortunately, when we activate it, server performance drops to about 1/10th of what it is without. After going through the handler and commenting out parts and benchmarking (rinse, repeat), Apache::Session appears to be the culprit. I can't believe that the use of this module should have so great an effect. We have to be doing something incorrectly. Any advice would be greatly appreciated. You are using Apache::DBI, right? Going to the database on every single request is going to be expensive. If the rest of your application doesn't normally hit the database it makes sense for this to have a significant impact. - Perrin
