JBoss - Weblogic comparison

[jojojv]

We are trying to do a performance comparison between Weblogic and JBoss application servers.
In the test scenario, we are using a Stateless Session Bean to insert a row into db and update the same with an Entity Bean.

With the same setup Weblogic has given 500 TPS while JBoss has given only 125.
I want to ensure that I have tuned JBoss server correctly.

Following are the parameters given for the test.

Execute Threads - 50 (Modified MaxPoolSize parameter for PooledInvoker mbean in jboss-service.xml)
Socket Readers - 17 (Modified NumAcceptThreads parameter for PooledInvoker mbean in jboss-service.xml)
Initial Connection Pool Size - 30 (Modified min-pool-size in oracle-ds.xml)
Initial Connection Pool Size - 50 (Modified max-pool-size in oracle-ds.xml)
Prepared Statement Cache - 10 (Modified prepared-statement-cache-size in oracle-ds.xml)
Initial Entity Pool Size - 500 Modified in standardjboss.xml
Max Entity Pool Size - 500 (Modified in standardjboss.xml)

I want to know whether I have modified the correct files. Also should I give any other additional parameters for improving JBoss performance.

Please help,
Jojo

[zdravko]

Hi,

We had a similar problem in our company, when tested JBoss3.2.5 and BEA's Weblogic 8.1 (SP1). Both are installed on Win2000 advanced server.

Web application on clients site do some upload. Problem is that using a JBoss app server for upload 5MB file takes ~130 sec, and when tried Weblogic with same file its takes only ~10-15 sec.

I couldn't explained that myself, and my colegues had no ideas too. Can anybody tell me how could I do something on perfomance tuning (JBoss).

thx,
Zdravko

[robisz]

Hi jojo!

Some tips for tuning CMP entity beans:

- Mark getter methods as read-only (in your jboss.xml)
- Turn on cache (in standardjboss.xml set commit-option to "A")
for your container (Standard CMP 2.x EntityBean)
- Read this article: http://www.onjava.com/pub/a/onjava/2003/05/28/jboss_optimization.html

[chrisdutz]

Another tip:
Increase the prepared-statment-cache for your datasource. This lead to a dramatic performance increas when doing some SPECjAppServer2004 benchmarking.

Yet another Tip:
Use the JRockit VM from BEA it did not increase the troughput much, but the response-time was a lot better. It's free and NO I do not get payed for recomending it ... If you have a look at the SPECjAppServer results posted on the web you can see that most vendors use JRockit in their benchmarks. Even If they could youse their own VMs. Don't think they do this because it suxx.