Hello Jonas,
Jonas Larsson wrote:
> I've done some very simple performance tests invoking methods of a simple
> session EJB on both ejboss 1.0 and orionserver 0.9 (a time ago;).
>
> The result was that orionserver was about 10 times - 50 times faster than
> ejboss. Both in VM calls and out of vm calls (both on the same machine).
>
> Can anyone confirm?
>
I don't recollect our actual numbers, but yeah, Orion should be substantially
faster. In our own testing we have yet to find a server that performs better
than Orion, but of course that is no proof that Orion is the fastest.
>
> The ejboss container uses the new dynamic proxy feature in jdk1.3
> (java.lang.reflect.Proxy etc) which I don't belive can be as fast as
> precompiled stubs because of the code overhead in the invocationhandler. The
> design and implementation with dynamic proxy is tough much simpler to
> implement. This might be the reason of the performance difference between
> ejboss and orion. (Orion team: Deny if I'm wrong!)
Actually, if JDK 1.3 is installed, we use dynamic proxies aswell, since it's a
very nice design. We even used this before they did. Are you sure they use it in
1.0? I thought it was introduced in the 2.0 alphas?
If you are using 1.2 we obviously can't use the dynamic proxies, but use an
approach where we compile the stubs on demand, transparently to the user.
So that is not the reason, however, we've worked very hard to design everything
in the server to perform and scale very well.
I am sorry for not being able to produce any benchmarks test numbers right now.
We don't want to publish results that we haven't veified since benchmarks is
such a touchy subject. But we do intend to publish benchmarks, althogh it's not
at the top of our priortity list right now.
Regards,
Karl Avedal
