Hi,
Think I am onto something:
For non-optimized (ie. remote calls)
within the server VM, the deserialization
(MarshalledObject.get()) of the
RemoteMethodInvocation does a class load
on *every* call. The class loaded is the
class I use for TPC: org.jboss.tm.GlobalId.
In the test case I set up for this, I call
from one entity bean instance in a container
to another bean instance in the same
container. This container's configuration
is set to use non-optimized calls.
The current performance for this is similar
to the 78 ms that Hugo reported in the
start of this mail thread.
So it looks like this *is* the JTA changes
i did.
But I don't really understand this: The
TransactionImpl we used instead before
this change deserialized fine.
And with the MarshalledObject, I
understand that we should have no
classloader issues, and that the class
should be loaded only once.
If anybody has an idea about what is
going on here, please let me know.
Until then, I'll try to find out some
way to make this problem go away.
Best Regards,
Ole Husgaard.
marc fleury wrote:
>
> |Hi,
> |
> |Could still be the embedded Tomcat
> |working in some way I did not
> |anticipate.
>
> could be... it's strange but real :(
>
> marc
>
> |
> |Just a quick thought: Could this be
> |because the embedded Tomcat is using
> |an old version of jboss-client.jar?
> |The changes I made affect the client
> |side too, so all copies of the JBoss
> |client jar needs to be updated.
> |
> |
> |Best Regards,
> |
> |Ole Husgaard.
> |
> |
> |marc fleury wrote:
> |>
> |> |The jbosstest load test gives almost the
> |> |same results with and without the JTA
> |> |changes.
> |>
> |> So it is the other stuff? the deployer changes from dewayne?
> |these were done
> |> at night 9pm so I don't think it is that...
> |>
> |> ok ok we will need to sit down... pfffffffff just no time...
> |>
> |> marc
> |>
> |> |
> |> |
> |> |Best Regards,
> |> |
> |> |Ole Husgaard.
> |> |
> |
