RE: New System Development Tomcat Tuning

Joe,

What version of Tomcat? I use 4.0.3. We have significantly more
than 20 users and never see the kind of response that you are
experiencing. I built my own startup scripts that give Tomcat
the memory it needs to run -- what values did you use? Do you
use native access? I am running Tomcat standalone, which is
supposed to run slower than IBM's installation but I wouldn't bet
on it (what do they have to gain by optimizing Tomcat?).

How hard would it be for me to duplicate your test on my system?

David Morris

>>> joepluta@PlutaBrothers.com 05/10/02 09:54AM >>>
> From: David Morris
>
> I get great performance from Tomcat. I can't tell that servlet based
> JSP applications are any slower than green screen. On a 730 with
> a CPW of 560/560, we get < .5 second response in either environment.
> It runs at an average of 60% utilization through the day taken at 30
> minute
> intervals. When I run similar code on a 270 with a CPW of 150, I
still
> get <
> 1 second response, where WebSphere ran about 4 (that was with the
> 1.2 JDK, and 3.5.2 - things may have improved).

David, I just ran some tests here on my little model 270.  Tomcat
performed
about 20-30% worse than WebSphere with my software under light load,
and
significantly poorer (200-300%) under heavy load.  This is at V4R5 with
all
current PTFs, and the Tomcat JAR files compiled as shipped at
optimization
level 30.  I'll rerun tests a little later at optimization 40, but I
don't
see that it could possibly reduce runtime to one-third or less.

My tests were not particularly scientific, but they were run on a
dedicxated
machine, and they jive with what IBM says about Tomcat as well.  Tomcat
runs
nicely under small loads but does not scale very well.  With a thin
servlet
tier such as the one I use, WebSphere clearly outperforms Tomcat.

The numbers will be in my WebSphere column on MCPressOnline coming out
Monday, but as a preliminary glimpse, on my 370CPW model 270, running
3
users simultaneously, with a user delay of 500ms and a page size of
about
10KB, I get an average response time of 131ms for WebSphere and 170ms
for
Tomcat.  Ramping that up to 20 users, WebSphere increases to an average
of
1040ms while Tomcat jumps to 3482ms.

This is just one man's observation, obviously, but it was consistent
throughout my tests.  I'll do some additional tests at optimization
level 40
when I get a chance.  Do you happen to know for sure which JAR files
need to
be recompiled?  From what I can tell, they are as follows:

 /QIBM/ProdData/HTTPA/admin/pgm/webserver.jar
 /QIBM/ProdData/HTTPA/admin/pgm/servlet.jar
 /QIBM/ProdData/HTTPA/admin/pgm/parser.jar
 /QIBM/ProdData/HTTPA/admin/pgm/jaxp.jar
 /QIBM/ProdData/HTTPA/admin/pgm/jasper.jar
 /QIBM/ProdData/java400/jdk12/lib/tools.jar

Joe