Re: [JBoss-user] Execution threads in Jboss

[R . Price]

>> Greetings all!  Can anyone tell me about how the execution threads work
>> between JBoss and Tomcat, and how many there are, can we change them,
>> etc.?
>>
>> We've put a slew of timing debug statements in our code to try to narrow
>> down
>> a perceived bottle-neck in JBoss.  We dump 100 virtual users on the box
>> and
>> see our execution time getting successively longer and longer.  It's
>> almost as
>> if there are only a couple of threads executing concurrently.
>Execution time gets longer _while_ the load ramps up or after it is
>ramped up?

Both.  You'd expect to see a load (to some extent) dumping so many in at
once, but once the virtual users are synched up and their transactions begin,
we see the load ramp up again until JBoss eventually implodes.

>>
>> For each user coming in, they hit a jsp page (which should be a unique
>> session
>> id for each user, which we have verified is so) which then is linked to a
>> stateful
>> session bean (which should also be a unique id for each user).  That EJB
>> then
>> uses a second stateful session EJB to make a telnet call out to a legacy
>> data source.
>Does this have to be stateful? Where is it removed?

Ooops.  The second EJB used is a Stateless Session EJB, not Stateful.  The first
has to be stateful because it retains information retrieved from the second EJB.

>>
>> That call from the second EJB to the data source takes 219 milliseconds.  
>Is this consistent?

Yes.  Always within 100-300 milliseconds.

>> When it
>> returns, the first EJB does some processing on it (which involves string
>> tokenizing
>> the results and stuffing it into vectors...blah blah).  When that returns
>> back to the
>> servlet class, we are seeing the time becoming longer and longer.
>The time between the completion of the call to the second bean an the
>return, or ???

This is the time of execution from when the call to the second EJB returns,
the processing internal to the first bean completes, and is then returned back
to the servlet behind the jsp.

When we've measured the middle part (processing internal to the first EJB), it
has been consistent.  Which is why that leads us to believe that there is
something being a bottle-neck between JBoss and Tomcat.  Maybe so, maybe not.

Robert