I did some performance testing using two machines which are dual
pentium4 zeons each with 4GB of memory, u320 scsi etc.

I used jakarta jmeter on one, and tested apache 1.3.28 - modjk2 - tomcat
5.0.16 - jetspeed1.4

the results are interesting. I can't say they're perfect - far from it -
because I ran each test only one unless something made me suspicious
over results (for example ensuring no developers on either machine).

-----Forwarded Message-----

Here's the results of comparing performance of our.psineteurope.com when
running different web services.

Note that apache was the front end to tomcat, tomcat runs jsps directly,
or the jetspeed webapp. The jetspeed home page was tested, as well as a 
particular jetspeed portlet I wrote which simply pulls a few rows from
a mysql db (on another machine)to see how much the DB slows things down.

The following table is the page response time in milliseconds when 200
fetches are done sequentially by a varying number of parallel users. Due
to the length of the tests.

(view in fixed-width font like courier-new)

Server| Users   1       10      33      100     500
---------------------------------------------------
Apache          1       1       3       3       4
Tomcat/JSP      5       22      25      19      32
Jetspeed        73      439     1448    4709    -
Jetspeed/mysql  60      432     1465    4787*   -


The results are the page response in milliseconds (1000th of a second).
I did not run a 500 "user" test against jetspeed, and truncated the 100
user jetspeed/mysql test before completion as it was taking so long, but 
it still gave useful results as confirmation.

Conclusions:

For very low loads, jetspeed runs adequately, but the performance falls 
off very badly as the number of parallel fetches increases. Apache, as 
would be expected, sustains a heavy load without flinching, and Tomcat/JSP
seems to cope well with high loads. 

The above figures are for serves running redhat7.3 and kernel 2.4.x, 
without tuning anything at all.
A 20% performance improvement can be almost guaranteed with a shift from 
kernel 2.4.x to 2.6.x.

Paul


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to