Re: CORBA performance - why CORBA

[Michael Meeks <michael ximian com>]

Hi Duncan,

On Mon, 10 Dec 2001, Duncan Grisby wrote:
> On Saturday 8 December, Michael Meeks wrote:
> > I'm not aware of an ORB where there are any issues with using
> > oneway in terms of loss / re-ordering.

	I stand corrected - it's great to have input from AT&T - may I
pop in next time I'm in the labs in Cambridge ?

> Any ORB that closes idle connections has an unavoidable race condition
> that can cause oneways to be lost. A server can choose to close a TCP
> connection at any time. Imagine a client fires off a oneway request,
> and while it is in flight, the server closes the connection. Some time
> later, the client sees that the connection has been closed, but it has
> no way of knowing whether the oneway was delivered or not.

	Certainly. This tends not to be a problem with Gnome because we
don't close idle connections. But true, with other ORBs it could be a
problem.

> With a normal two-way invocation it would have seen that the
> connection was closed before it got a reply, and would therefore know
> it had to retry with a new connection.

	Yes; not that we actualy re-try with a new connection either :-)

	Regards,

		Michael.

-- 
 mmeeks gnu org  <><, Pseudo Engineer, itinerant idiot