[omniORB] Long remote call processing

[Duncan Grisby]

On Wednesday 29 November, radamkie at kdm.pl wrote:

> Here is some parameters of configuration that was set in an application
> client(JaCORB)-server(omniORB)
> 
> ORBendPoint=giop:tcp::45016
> ORBdumpConfiguration=1
> ORBclientCallTimeOutPeriod=10
> ORBmaxGIOPConnectionPerServer=20
> ORBmaxServerThreadPerConnection=1000
> ORBoneCallPerConnection=0
> ORBserverCallTimeOutPeriod=10
> ORBconnectionWatchImmediate=1

Why did you choose those parameters?  They are a fairly arbitrary bunch.
What happens if you use the default omniORB configuration without
changing any settings?

> Client send circa 20-40 remote method calls per second to circa 300 CORBA
> different objects resisted in server. It's measured by client side a time 
> between Start and End of executing remote method call (result A), and also
> by server side a time between executing first line of remote method and
> last line(result B).
> 
> Generally, results show that A < 3 ms  and B < 1 ms  but few times(4-10)
> in a minute, result A has a great peak, circa 100-300ms (result B is still
> < 1ms).

How many threads does the client have?  If more than one, Is JacORB
configured to multiplex calls on one connection or to open multiple
connections?

Are any of your operations oneway?

If you are using omniORB 4.1.0 (or can upgrade), run with -ORBtraceLevel
25 -ORBtraceInvocations 1 -ORBtraceThreadId 1 -ORBtraceTime 1 to see if
that reveals anything useful. With earlier releases you can use all the
options other than -ORBtraceTime, but that will make it harder to follow
what's going on.

Cheers,

Duncan.

-- 
 -- Duncan Grisby         --
  -- duncan at grisby.org     --
   -- http://www.grisby.org --