FWIW, I did a trivial benchmark of Jonas vs. Weblogic.
The results may be of interest. I think it's way less
robust and exhaustive than the ones already discussed,
but more fair (comparing apples to apples). I'd really
love to hear from anyone that has more data to add to
this discussion even if it has been played out on the
mailing list.
We're admittedly using a very small subset of the features
of an EJB application server (there is internally some debate
whether we should use one at all). We can, without
too much trouble, avoid the use of entity beans altogether.
What we hope to get from using the server is to
implement a n-tiered architecture where many clients connect
to the middle tier which is serving out the classes that
contain our business logic and connect to the back end.
We could do this in other ways, but (please correct me if
I'm wrong) an EJB server saves us the trouble of writing the
code to instantiate the classes, manage the threads, etc.
Not a big deal, but I'd rather let someone else do it ;-)
Most of our beans will be simple queries that will not require
any kind of persistence in the database. Those that do
we will probably code up ourselves rather than relying on
CMP.
So, here's what the test looked like:
I have a "hello world" servlet running that connects to
the weblogic server and does 10,000 calls (essentially
no-ops) on a stateless session bean. The reference to the
bean is obtained once in the init procedure of the servlet.
In effect, I wanted to know the absolute minimum cost of
a remote method call on a stateless session bean. This actually
resembles our target application.
The test was run on a single 500Mhz pentium, 128M ram,
linux kernel 2.2.14-15mdk, running Apache 1.3.9, apache
jserv 1.1, Blackdown JDK 1.2.2 RC4. Everything runs on
the same box. There is no dbms access.
I ran it using Weblogic 5.0 beta and Jonas 1.6.
To execute 10,000 calls that essentially marshall a double
the Weblogic server required ~42 seconds and Jonas ~11 seconds.
The really interesting difference was in the resident memory
segment size of the java threads run by the ejb server. They
went up to 72M for weblogic! and were negligible for Jonas
threads (the blackdown jdk 1.2.2 is using native linux threads
I believe).
The cost of Weblogic licences is a serious problem for us.
We're quite happy to scale the app by throwing multiple
EJB server boxes at it (and use JNDI & LDAP to pick a server
at random) but at $15K US per box, we're going to have to
redesign our app to use fewer boxes, which is by most accounts
a _very_bad_thing_to_do_.
--
http://www.singingfish.com
(206) 691-0565 ext. 3000
----
To unsubscribe, send email to [EMAIL PROTECTED] and
include in the body of the message "unsubscribe jonas-users".
For general help, send email to [EMAIL PROTECTED] and
include in the body of the message "help".
