Philippe,
> [EMAIL PROTECTED] wrote:
> >
> > We tried Jeremie as well. The problem is that Jeremie speeds up, but still it is
>so slow that I cannot believe that there is no bug. With RMI we have a >
rate
> > of 45 method calls per second. With Jeremie, there are about 70. This is near to
>200%, but still this is quite slow. I think there is either a bug in >
JOnAS
> > (can someone give me a tip where to look in the code?) or I have done something
>wrong. Maybe I have to register the entity bean as a reference to the >
session
> > bean or something, so that JOnAS knows that both are located in the same VM?
>
> I don't think JOnAS is particularly slow, comparatively to other Application Servers.
> I think that simply using EJBs instead of direct code has a cost. For example, every
> finder method will search in the database (it's in the EJB spec) for the bean
>instance.
> USing Jeremie in place of RMI will just avoid overhead due to bean-to-bean
>communications
> when both beans are in the same JVM, but it will not avoid database accesses for
>example.
Hm, well, I will not insist on there is a bug in JOnAS, but I cannot believe that
Entity Beans have such a huge overhead compared to Session Beans with JDBC.
I'm more or less sure that I did a failure somewhere, but I do not know why. Maybe
there is a mistake in my DD or something, but it looks quite well.
My problem is, and this is the point where I cannot believe that EJB is the problem: I
can load about 1000 records by JDBC into a stand alone JDBC
application, but I can load only 39 Entity Beans (CMP, via Session Beans's method with
"RequiresNew"). Transaction overhead is not the problem (we did a test
with and without TX). Maybe I should post a small example, sure a professional will
see the problem? Another phenomenon is that our measured 39 EB loads
split linear with increasing client count, that means, if I am using 4 client
workstations loading DIFFERENT (!) Beans, I have 39 divided by 4 EB loads per
second per client. I would commit to 39 divided by 4 is correct behaviour if loading
the SAME bean from every client, but for several reasons
(multi-threading etc.), with four clients loading four different beans (4 different
data rows on DB), performance should be at least a small bit faster than
39 divided by four, since there is no locking etc.
thank you
MArkus
