Re: Re: JOnAS 5.2 Benchmarks available?

Guillaume,

I did a short test on our current JOnAS 5.2 project:

1 SLSB method with required-TX calls another SLSB's local-interace
method (also with required-tx ) within a loop 1000 times.

the 1000 times called method does a very simply:
entityManager.find(class,id) call and gets a JPA entity (that already is
in the second level cache)

I expected this test to run probably some seconds but the result was:

it took over 60 seconds --- so this means that a single  call of a local
SLSB method which does a simple em.find() takes about 60 milliseconds!!!

I did not have the time to set up a project for JOnAS4 to make a
comparable run but I think you will agree that doing a single local
interface method call
which does a single em.find (always the same object ID - with activated
second level cache) --- 60 milliseconds is very very slow for this
operation ! 


we made other obersavtions on our production servers - this is not
exactly comparable but there is 1 instance left running the old JOnAS4
based system:
the processing of a submitted datapoint of a telemetry system takes
about 50-60 milliseconds there
on the new servers we observed 100-200 milliseconds .

I'm wondering nobody else observed/mentioned performance issues of EJB
method calls ? 


I know that doing remote interface method calls are SLOW but from my
experience with JOnAS4 - local SLSB method calls have been really fast
-- seems that
this behaviour changed on EZB ?


Any ideas how we could improve performance ?? (configuration?)


regards,
HANS

Am Montag, den 28.03.2011, 15:42 +0200 schrieb Guillaume Sauthier:

> 
> 
> Le 28/03/2011 10:21, Hans J. Prueller a écrit : 
> 
> > Hi all,
> > 
> > just for my interest - are JOnAS 5(.2) benchmarks e.g. compared to
> > JOnAS 4.x available?
> 
> 
> AFAIK, no
> But if you have measurements that you want to share ...
> 
> 
> > 
> > Basically JOnAS 5.2 runs smoothly but in some cases it "feels"
> > tough/slow - from a quick sourcecode analysis
> > it seems to me, that EJB calls  (SLSB)  are slower or more costly
> > now than in JOnAS  4 - is that true ?
> 
> 
> At least, we can say that the technologies used to implements EJB3 are
> very different of the ones used for EJB2...
> But, in all the cases no Java reflexion is used on the server side
> (this is usually the costly thing).
> 
> On the client side, this is a Java proxy (reflexion) that is
> registered on JNDI for EJB3, so here we may have a performance
> penality ...
> In EJB2, the stub was directly registered in JNDI (no reflexion)...
> 
> Hope it helps...
> --G
> 
> 
> > 
> > regards,
> > hans
> >