[omniORB] object deletion strategies

David Riddoch djr@uk.research.att.com
Wed, 19 Apr 2000 15:07:36 +0100 (BST)


On Wed, 19 Apr 2000, Stefan Seefeld wrote:

> David Riddoch wrote:
> 
> > Bear in mind that omniORB 3 _is_ the future of omniORB, and that is where
> > the development effort is going.
> 
> Sure. I'm just wondering whether the POA is really the best object adaptor
> we can get in the particular context of berlin. It is certainly a major step
> ahead for 'normal' CORBA applications where granularity of objects is quite high.
> In Berlin I do lots of scene graph traversals per second and each involves a
> high number of method invocations, even a couple of transient objects 
> (transformations and regions notably for dynamic layout).

I appreciate the problem.  Just so there is no confusion, in omniORB 3 the
object implementation and reference are separate even if you use the BOA.
There is no performance advantage in using the BOA.

Supporting object references which are direct pointers to the
implementation cannot be supported without major effort, would be
non-standard and would probably impact the performance for the normal
case.  I'd guess that it is unlikely to happen...


> Berlin is currently quite slow, Fresco is much faster. I'm wondering where
> the differences come from, given that Berlin inherits Fresco's architecture.
> One possible difference is that Fresco uses a non-standard ORB. That's why
> I'm trying to understand the particularities of all the involved
> strategy choices.

Odd.  Given that most of your calls are currently just virtual function
calls I would be very suprised if it is the ORB that makes the difference.
You can't get much faster than that!


Cheers,
David