We have integrated RMI version of Jonas with Jetty v 2.3.x (another free
Java only web server)and the RMI registry into one application and the
performance nicely improved. It is still not as fast as Weblogic, but it is
finally usable for our purposes. This even got us around the hardcoded (60s)
transaction timeout value in Jonas (I didn't get any response on my
question, if this will be customizable in Jonas).
Miroslav Halas
Software Engineer
Compuware Corp.
15305 Dallas Parkway
Suite 900
Addison, TX 75001
phone 9720-960-0960 x 1333
fax 972-960-8489
-----Original Message-----
From: Gerard Vandome [mailto:[EMAIL PROTECTED]]
Sent: Thursday, February 10, 2000 11:55 AM
To: Bolognese Luca
Cc: Robert Krger; [EMAIL PROTECTED]
Subject: Re: Jonas vs. Weblogic
Bolognese Luca wrote:
> We evalutated Jonas, Weblogic and a JSP only scenario.
>
> The test was 3 or 4 jsp just asking some data to some Session Stateless
EJB.
> These were connected to an Oracle instance running in another server.
>
> We tested it in NT, Linux and HP-AUX.
> We tested it using 10, 20 and 40 thread each one asking one jsp page at a
> distance of 3 seconds between each request.
>
> We used Apache 1.3.9 with JServ 1.0 and gnujsp 1.0 for JonAS, the
integrated
> web server of weblogic.
> Jdk was 1.1 because was the only one avaiable on HP-AUX 10.20.
>
> Our machine was a Pentium II with 128 M RAM, 1 processor.
>
> I can't report the results in full because I don't have them in electronic
> format yet, but in synthesis:
>
> Times are in milliseconds, cl stands for clients:
>
> Weblogic response time was: 123 with 10 cl, 150 with 20 cl, 220 with 40
cl,
> 650 with 60 cl
> JonAs 1.4.1 was: 725 with 10cl, 1200 with 20 cl, 4000 with 40 cl.
> Using normal Java classes without them being in a EJB container: 327 with
10
> cl, 370 with 20 cl, 660 with 60 cl, 1900 with 60 cl.
>
> The last row is probably because of the poor implementation of Apache on
NT,
> on Linux and HP-AUX Response time are the same as Weblogic until we get to
> 40 cl then they start to fall.
>
> This was on NT machines, performance on Linux and HP-AUX was a bit worse.
>
> The problem was all about %CPU , Jonas with only 20 cl takes 100% of CPU
> time.
> Profiling the application we saw that the problem was all in the RMI part,
> so essentially in the data passing
> between JServ and JonAS. When we used an EJB calling another one things
got
> worse then that, because there was more marshaling going on.
>
> We think that an integration with some web-server can make the situation
> better.
> This is only an our supposition, of course.
>
No you are fully right.
Your figures above don't compare the same thing, as in the case of Weblogic
your
are in the same JVM (local RMI call) while in the case of JOnAS you are
using
remote RMI calls.
The integration of a web server with JOnAS (using Jeremie) within the same
JVM
will drastically improve the figures for JOnAS: it will be several hundred
times
faster (Jeremie local calls instead of RMI remote calls). It is what we are
currently doing by integrating Enhydra with JOnAS allowing the possibility
to
run both within the same JVM. we expect a first version end of March.
Even without integration within the same JVM, the use of Jeremie with the
current version of Jonas will drastically improve performances in the same
way
when using an EJB calling other ones.
G�rard
G�rard Vandome BullSoftware
1, rue de Provence B.P. 208 - 38432 ECHIROLLES CEDEX - France
http://www.openmaster.com/ejb [EMAIL PROTECTED]
33 4 76 29.75.67 mobile 33 6 85 22 34 95 fax 33 4 76.29.76.00
----
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".
----
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".
