Re: Performance problems with mod_WebObjects.so, FastCGI

On Jul 14, 2004, at 7:14 AM, Jürgen Lorenz Simon wrote:

> Hi,
>
> I have a problem with performance of WebObjects Adaptors on SuSe Linux 
> 9.1
> using Apache2 and WebObjects 5.2.3.
>
> I compiled and used the mod_WebObjects.so and it takes twice the time 
> of the
> normal CGI-BIN adaptor.  Using the FastCGI adaptor results in a 
> slightly higher
> time than the CGI-BIN adaptor.
>
> So it seems the CGI-BIN adaptor performs best of the three and that 
> seems very
> odd to me since that is not very fast to start with.
>
> Has anyone made similar experiences?
> Any known pitfalls I may have overlooked?
> Configuration problems to observe when using either of the two?

  Yeah lots.

  Ok, so first off, you're using Apache 2. So which MPM module you use 
makes a bit of difference, and then how Apache is configured after that 
makes some difference. After that, getting your config file correct for 
the adaptor correct is important, unless you're using Monitor which 
writes them for you, but has other issues.

  I haven't dived into the FastCGI adaptor, but here's an overview of 
why the normal cgi-bin should be slower.

     1. Using the regular cgi should require launching of separate 
process for each
     request. I think the regular cgi might use shared memory now, so it 
can
     round robin, but it didn't used to be able to. Though what happens 
to
     that shared memory when no cgi processes are running? Hmmm...

     2. Using mod_WebObjects.so shouldn't require launching processes, 
but should
      otherwise be running the same code as the cgi. Its possible that 
you're using
      the 1.x adaptor with Apache 2, which perhaps requires some sort of 
adaptor
      layer inside WebObjects if that's even possible.

  If you're seeing the cgi adaptor faster, then you have something weird 
going on.

  Possibilities:

      1. The CGI adaptor isn't working as hard as the regular adaptor. 
It might
      just be picking the first instance in the list (which happens to 
be running)
      and quitting. The module might be looking at all the instances 
that are
      supposed to be running, doing a round robin to one of them (which 
happens to
      not be running), then failing over.

      2. As I said, perhaps you're using a 1.x adaptor with 2.x. The 
Project
       Wonder adaptor was just a quick port of 1.x to 2.x last time I 
looked,
       its possible they missed something subtle. I know they didn't do 
anything
       with regards to the multiple MPM models in Apache, which is 
disappointing.

      3. Perhaps there's something weird on SuSe where loading a shared 
library
         takes longer then launching a process. That shouldn't matter, 
because
         your MPM for apache should be setup so that it should only have 
to do
         that when it launches a fresh apache instance, but the setup for
         apache out of the box isn't that good.

      4. There's something wrong with how you're doing the benchmark. 
What are
        you using?

  So there's lots of stuff to look at there, but I'm not sure anyone can 
be much help with the information you've provided so far.

   Remove WebObjects from the equation and try benchmarking the sample 
CGIs that come with apache against the fast-cgi versions. If fast CGI 
isn't faster, you've got something else wrong.

  Pierce