Is there any performance comparison between Perl w

问题:

I have seen mentions (which sounded like unsubstantiated opinions, and dated ones at that) that Embperl is the fastest Perl web framework.

I was wondering if there's a consensus on the relative speed of the major stable Perl web frameworks, or ideally, some sort of fact-based performance comparisons between implementations of the same sample webapps, or individual functionalities (e.g. session handling or form data processing), etc...?

UPDATE: This question is specifically about the speed comparison of different frameworks, executing identical/equivalent tasks. I appreciate the good intentions, but I already know that speed is not the only criteria I should be looking at. I wasn't asking for philosophical advice. And believe it or not, being frameworks, you CAN actually compare their speed on an apple-to-apple basis by running identically purposed tasks/code/apps on them (e.g. render a given form with a given set of templated inserts etc...), even if the full functionality of each framework is not 100% the same.

回答1:

Here is one comparison between perl frameworks, in terms of speed (startup) and memory consumed by framework itself. It is a bit old (2008), so it does not compare new stuff like Plack.

http://mark.stosberg.com/blog/2008/11/startup-benchmarks-for-mojo-catalyst-titanium-httpengine-and-cgiapplication.html

回答2:

I don't want to go into the interpretation discussion (for most real world scenarios these overheads have no impact at all) - but here are my tests:

1. Pure Plack

zby@zby:~/progs/bench$ cat app.psgi 

sub {
   my ( $env ) = @_;
   return [
       200,
       [ 'Content-Type' => 'text/text' ],
       [ 'Hello World' ]
       ];
}
zby@zby:~/progs/bench$ plackup
HTTP::Server::PSGI: Accepting connections at http://0:5000/

With simple ab -n 10000 I get

Requests per second: 2168.05 [#/sec] (mean)

2. Dancer

zby@zby:~/progs/bench$ cat dancer.pl 
 #!/usr/bin/perl
           use Dancer;

           get '/' => sub {
               return "Why, hello there";
           };

           dance;
zby@zby:~/progs/bench$ perl dancer.pl 
>> Dancer server 1950 listening on http://0.0.0.0:3000
== Entering the development dance floor ...

With similar ab test I get Requests per second: 1570.49 [#/sec] (mean)

3. Mojolicious::Lite

zby@zby:~/progs/bench$ cat mojo.pl 
 # Using Mojolicious::Lite will enable "strict" and "warnings"
    use Mojolicious::Lite;

    # Route with placeholder
    get '/' => sub {
        my $self = shift;
        $self->render(text => "Hello!");
    };

    # Start the Mojolicious command system
    app->start;
zby@zby:~/progs/bench$ perl mojo.pl daemon
Sat Jan 22 20:37:01 2011 info Mojo::Server::Daemon:320 [2315]: Server listening (http://*:3000)
Server available at http://*:3000.

Result: Requests per second: 763.72 [#/sec] (mean)

4. Catalyst.

Unfortunately the code is much too long to be presented here in its entirety, but the Root controller contains:

sub index :Path :Args(0) {
    my ( $self, $c ) = @_;

    # Hello World
    $c->response->body( 'Hello World' );
}

The result is:

Requests per second: 727.93 [#/sec] (mean)

5. WebNano

zby@zby:~/progs/bench$ cat webnano.psgi

{
    package MyApp;
    use base 'WebNano';
    1;
}

{
    package MyApp::Controller;
    use base 'WebNano::Controller';

    sub index_action {
        my $self = shift;
        return 'This is my home';
    }
    1;
}    
MyApp->new()->psgi_callback;
zby@zby:~/progs/bench$ plackup webnano.psgi 
HTTP::Server::PSGI: Accepting connections at http://0:5000/

And the result:

Requests per second: 1884.54 [#/sec] (mean)

This will change after some more features are added.

回答3:

I know this doesn't answer you directly but I don't think an up to date comparison exists and I know a comprehensive one doesn't. It would take a couple weeks of work, at least, to do a thorough benchmark because there are so many frameworks in Perl right now with so many DB/Template/Server permutations and different kinds of usage patterns of the app could make major performance differences too.

I do believe you will be missing a lot by taking Mark's simple 2008 benchmarks as the answer to your quest. Deployment matters as much as if not more than the web framework for speed. For example, Catalyst is not going to win a raw "hello world" speed war but the BBC's video Catalyst application can serve 1,000 concurrent videos. Flexibility, scalability, and support of different deployments becomes a big factor in picking a web framework.

Plack is new and major. In just one year it's seen a huge adoption, middleware/plugin growth, and support from just about every framework. The Starman engine for plack apps is surprisingly fast and supports hot reloads and worker process increment/decrement. Since almost all the Perl frameworks can run as .psgi now you could run whatever you want on Starman + nginx (or lighttpd). There are dozens of good deployment combinations and quite a few changes and new entries in the web framework space in the last two years.

If you're doing modern web stuff, make sure to pick a kit with websocket support. That alone will increase performance dramatically over traditional Ajax; small requests/responses can be a factor of 100 times smaller/lighter with websockets.

Sidenote: modperl is probably not the best persistent deployment to pick at this point unless you have a need for the deep hooks into the request cycle. It has many caveats and wrinkles and ties you to apache (a great server but not the fastest option by a long shot).

Happy hunting!

Update 20 October 2016: uWSGI is a fantastic match for PSGI apps in Perl.

回答4:

TechEmpower have a some good comparisons of framework performace. See http://www.techempower.com/benchmarks/

As of now, they have included 4 perl frameowkrs (Dancer, Kelp, Mojolicious, Web::Simple) as well as many frameworks from other languages. The results can be filtered to show only Perl frameworks if desired.