JBoss 4.0.2 versus IBM WAS5.1.0.4

[caseswitch]

hi,

we are relative new to jboss - just testing the migration from IBM WebSphere Application Server 5.1.0.4 to JBoss 4.0.2.
Today we tested the performance between the two application servers...with no good result for JBoss, i think! :(

Our systems:
-Linux/s390 (kernel 2.4) running under a zVM5.1 in an IBM z900 zSeries beast (production running z990).
-IBM HTTP Server 1.3.28.1
-IBM DB2 8.1
-IBM CICS Transaction Gateway 5.1 (using IBM Java131 SDK)

Our Application:
It's a webfrontend, replacing the simple 3270 emulation (HOST) and using a DB2-Enterprise located at our mainframe. Connection to CICS is done via the CICS-gateway (haha).

Server 1 runs JBoss 4.0.2 (with IBM SDK 142), server 2 runs IBM WAS5.1.0.4.
Testtool is the LoadRunner/WinRunner Suite by Mercury.

In first run, the Servers were virtually equipped with "Hardware" configured as:
-server1(jboss): 1024MB, 1 VCPU
-server2(WAS5.1) : 768MB, 1VCPU.
WAS5.1 JVM Heap was min256 and max512 - same for JBoss => the winner was...IBM WebSphere
Second run i reserved more memory for server1 within the zVM. Pushing it up to 1280MB. The winner was...IBM WebSphere.
In the third run i changed the JVM max. heapsize of JBoss to 768M...the winner was WAS5.1 again but JBoss performance increased a bit.

It can't be a result of HTTPD<->JBoss - when testing with JBoss' native 8080 the performance decreases...another bit slower.
It can't be the VCPU - zVM is running with 3 dedicated Linux-IFLs. Their load when benchmark is running is round about 20% each...no problem here!

Question is: where to look for? Or: Does IBM WebSphere always perform better than JBoss?

[clebert.suconic]

It depends on how you configure/use JBoss.

Also, are you using CMP.

What are you doing?

What tests are you using with Loadrunner? (Did you create your own tests?)

[caseswitch]

hi,

thanks for your reply.

no - we aren't using CMP/EJBs.
forgot to say in the leading post: i'm a sysadmin - i'm *not* a java/webapp developer.

what we are doing:
loadrunner-scripts are done by our developers. better: the script is a recording of a simple end-user action our employees would to.

-searching DB2 tables located at our mainframe (DB2)
-CICS transactions

the test-script doesn't change anything. the recording is a simple reading task.

1) searching a customer via name, customerid or anything else (CICS-transaction)
2) viewing the customers data (CICS-transaction/DB2)
3) do another search
4) view data...
5) and so on.

the application is *far* more complex, but the test itself is quiet as simple as written above. point is: the server hasn't got much to do. test runs in T-environment. DB2 responses and CICS-transactions running at normal speed - the server hasn't to wait long for other systems to deliver it's data.

DB2 stuff is done via DCS-connection (DB2 connect on webserver) to our mainframe.

regards,
-Case

[zzy9v4]

I've tried to produce a 1166 pages PDF with iText 1.4 and while WAS 5.1.1.9 take 90 sec, JBOSS 4.1 take only 25 secs.
But platform is a Laptop with Windows XP sp2 with 1Gb of memory and a Centrino 1,6 Ghz.