>
> I do think we need to invest some effort on the protocol level that can signal
> application programmer that data flow has/has not finished.
>
> As for callback function triggered by monitorOn, I do not think this is a
> problem we have to worry about. However I think the real problem resides
> in the cases invovling transfering large amount of data (say ~1 mbytes)
> in a single cdev call but the server sends data in separate packets. Ideas :-).
>
> --Jie Chen
Well, the monitorOn thing was only one example... the real problem
that I am trying to address is in the case of a CDEV Gateway. The
Gateway is responsible for creating transaction objects to send
messages to CDEV devices that it knows only by name. How does this
gateway service know that the response that it received from the
gateway server is the last one that it will receive in response to
that transaction. How does the gateway service know when it can delete
the transaction object that it created to track the message...
The current idea is that the gateway server will include a "complete/
incomplete" indicator in the "op-code" component of the packet - BUT...
in order to do that, the server has to know when it gets the callback
from the "real" CDEV service that it is either finished or not
finished...
The server must either have a generic way of finding out that the
message is really finished - or - it will have to know EVERY message
and whether or not it is a multipart transaction...
Walt
--
=============================================================
Walt Akers Voice: (757)269-7669 E-Mail: akers@cebaf.gov
Thomas Jefferson National Accelerator Facility
12000 Jefferson Avenue, MS 16A
Newport News, Va 23606
=============================================================