April 1, 1997 Meeting Minutes

Actual Minutes

Participants

Chris Cleeland, Brian Mendel, [briefly] Andy Gokhale and Doug Schmidt

What Happened

• Doug and Chris discussed the current step for porting more of the ORB to use ACE components.
• Andy had nothing new to report on performance figures.
• Brian and Chris discussed demultiplexing strategies, reviewed Doug & Chris' discussion of ORB->ACE porting step, and lamented having to rewrite the Object Adapter.

Resolutions

• For using more of the ACE components, the current step is divided into two smaller steps. The first uses ACE's IPC_SAP components in place of socket-related ACE_OS calls. The next step replaces the connection management substrate with the archetypical Service Manager architecture (need picture!).
• In consideration that multiple demux stragies exist and we might want use any of several of them, we discussed how to make the selection of strategy flexible. But, in order to do this we will implement the simplest strategy first (likely the static one).

Open Issues

• Must all review the CORBA2 spec to understand the desired Object Adapter interface so that we are proceeding towards a common target.
• We still don't know where the other 12Mbps of performance got lost.

Miscellaneous Notes

We identified the following demultiplexing strategies:

• Static, where something, presumably the IDL compiler, generates a substantial amount of demux'ing information prior to runtime.
• Dynamic, which requires dispatch to a virtual function on CORBA::Object, e.g., handle_request(ServerRequest& r). This method would, after examination of r, forward to the appropriate method on the real object.