CORBA connections not relinquished - CPU usage - 100%

Question

We have a session bean making a corba call, however after the call is susccessful,
the connection doesn't relinquished. If We do a netstat , the number of established
connections keep increasing as more calls are made to the CORBA server. We am
facing issues of CPU usage goin upto 100% and never comes below 97% after a lot
of such calls. Please suggest someway of fixing it.

Answer 1

You should verify it within the product documentation.
If you want to verify it by yourself, from wls, implementing the singleton pattern
for your remote object, invoke a method on the remote object, implement a sleep
in it (on the corba server), and after try to execute another method on the same
remote object (singleton), which returns immediately (it is important that you
run this test within the same wls instance, and the same ejb (ejb module) : the
singleton is relative to the classloader).
If the orb implementation does not multiplex the invokations, you will see the
second call not returning until the first one has not returned (end of sleep time).
(of course, I suppose the Corba server is multithreaded...)
Sergi
"Nishant" <[email protected]> wrote:
>
Perfect....thanks....but isnt using singleton going to harm the performance..coz
this is old version of visibroker, and it prbably doesnt multiplex the
threads??
"Sergi Vaz" <[email protected]> wrote:
Hello,
use the ORB as a "singleton" object : do not invoke ORB.init many times,
but only
once, and after cache the corba object on a static variable.
It's thread safe, the remote invocations will be multiplexed.
So you will have only "one" reader thread for each corba object, and
you will
solve your problem.
Some vendor are not fully compliant and do not implement the destroy
method.
So you do not release the ORB resources, but keep them (it's also more
performant).
At this point you will have a "dangling" reader thread only after aredeploy
(unless
you shut down the jvm).
Sergi
"Nishant" <[email protected]> wrote:
Andy,
I am using visibroker 3.2. , and WLS 7.0 , JSK 1.3.1_08 And I do not
see a method
called org.omg.CORBA.ORB.destroy.
Is there an alternative way to achieve this??
Unless I destroy the threads the connection with the CORBA server would
remain,
and this is driving up the CPU usage, for every new connection made.
Thanks
Nishant
Andy Piper <[email protected]> wrote:
"Nishant" <[email protected]> writes:
We have a session bean making a corba call, however after the callis susccessful,
the connection doesn't relinquished. If We do a netstat , the numberof established
connections keep increasing as more calls are made to the CORBA
server.
We am
facing issues of CPU usage goin upto 100% and never comes below
97%
after a lot
of such calls. Please suggest someway of fixing it.This is most likely because you are using the JDK 1.3 ORB. It doesnot
release threads until the system exits. This is one reason why using
the BEA ORB in 8.1 is preferrable.
If you are using WLS 7.0 then Sun did make a fix in JDK 1.3.1_09 which
adds a special property to allow ORB.destroy() to release the
threads. You would have to go through support to find out what this
property is. If you are using 8.1 then you can use the BEA ORB or
if
you need to use the JDK ORB I think that calling ORB.destroy() will
release the threads for you.
Hope this helps.
andy