Mark Guzdial wrote:
> The documentation (PDF) is very nice, and includes a use of the suite
> to compare Comanche and Apache. (Comanche performs at about 30% of
> Apache, overall.)
Some thoughts on this...
I'm interested in knowing if Squeak can be used as platform for developing
commercial sites. My major concern about this is performance, we all [here]
know that Squeak is terrific for building applications, that's not the point.
I started to think a some time ago, what test should one perform to solve
this problem (to know if squeak is finally ready for such task).
One of that ideas I had is comparing Apache+perl, Apache+mod_perl,
Apache+php3 and IIS+ASP [microsoft's stuff] with Comanche+Squeak:
perl, php3 and ASP are all interpreted languages (I believe), as well as
Squeak is.
when perl and php scripts are used for cgi-bins, every time somebody
activates one of this scripts, the interpreted is loaded, the script is
compiled and the it's ran, (unless you use mod_perl, and php3 as an Apache
module), I don't know what's about IIS+ASP. So I wanted to compare the
performance of all this mixes:
perl without mod_perl (just #!/bin/perl, sometimes used),
which reloads and interpreter and recompiles the script every time.
perl on mod_perl
php3 without Apache module (#!/bin/php3 not usually used)
which reloads and interpreter and recompiles the script every time.
php3 on an Apache module
IIS+ASP
Squeak+Comanche
Squeak+Comanche+FourStrongWings
Some thoughts on this too:
Squeak's interpreter (VM) is always on memory, the Scripts are always
compiled, Squeak's interpreter doesn't have a JIT, but it will have a good JIT
soon.
I can understand that Squeak's performance when transferring static files
is poor, I was expecting it. But I'm not sure what'll happen when comparing
dynamically generated pages.
This test are not easy to do, the compared applications should be similar
in complexity for the test to have really meaning (or not?).
And then, based on this results, I'm planning to set up a Linux machine,
with kHTTPd(*) and Squeak+Comanche, to test overall performance of the mix:
(*) kHTTPd is a kernel module for Linux, that serves HTTP for static files.
I believe (from what I read) that it can be configured to serve all the files
under a given path, and for HTTPRequests that fall outside that path, forward
the requests to a standard http server listening on port 80 (let's say
Comanche).
With this configuration (**) I'm hoping we can achieve a good overall
performance for using Squeak as a Commercial web server.
(**) This configuration is not enough, of course if you dynamically
generate every page in your site, there's no gain. A good design of the site is
needed: We need to minimize the amount of dynamic content, however, this
doesn't imply that the site will loose functionality, for example, on a swiki,
every page may be generated when somebody changes it, and saved in a static
file, this'll reduce the amount of dynamic content for searches, and histories
(that can also be pre-generated)...
Ok, this all popped out of mi mind right to the keyboard for the first time
now, so it may be all wrong...
Pre-Generated Bye!
Richie++
PS: I hope you understand may english... I never know if somebody else (but me)
will understand what I wrote...
