Re: Which scales better? J2EE, PHP, or CGIDEV2?

[Thorbjoern Ravn Andersen <ravn@xxxxxxxxxx>]

 Den 07/10/10 01.17, Nathan Andelin skrev:
> > From: Aaron Bartell
> > I would be curious to know how much CPU time is required to
> > create a new J2EE thread vs. have Apache hand-off a "CGI" request to
> > an existing job?
> You're probably dealing in microseconds either way; less than 1 millisecond.
>   But how often do JEE application servers create new threads?  The IBM i HTTP
> server appears to reuse threads activated at startup.  Do JEE servers create a
> new thread for each request?  Like they create new "request" and "response"
> objects for each request?
Usually not.  The typical Java EE server keeps a pool of worker threads 
which grows and shrink according to load.


> Actually, they say that JEE servers only create one instance of each Servlet.
That instance can be invoked from different contexts easily.  A servlet 
request do not keep state in the servlet object, but in servlet contexts 
provided as parameters.

Hence no need for more than a single object per servlet.

>   Servlet methods may be called by each thread.  That would be quite a bit
> different than CGI, where you might have an instance of each *PGM loaded in each
> CGI worker process.
>
> JEE app servers evidently synchronize Servlet method calls.  Servlet methods can
Only if explicitly requested for non-threadsafe code.  The default 
servlet model is unsynchronized.


> easily get quite complex.  That's probably why JEE app servers DON'T scale well.
>   Okay, I said it. IMHO, they don't scale well. Studies show that JEE application
> servers don't have the ability to effectively utilize multiple processors,
> unless you launch multiple instances, and deploy applications under each.
I would like to see your studies.  That experience is contrary to mine.

(but it would be consistent with a deployment with non-threadsafe servlets)