[Openais] flow control and merge recovery

[Mark Haverkamp]

On Fri, 2004-09-17 at 13:24, Steven Dake wrote:
> Mark & Sakai-san,
> 
> I have considered what happens during a merge when alot of data needs to
> be synchronized.  Right now, the gmi_mcast will eventually fail to queue
> messages because it runs out of buffer space.  (after 173 small
> messages, I think)
> 
> I was thinking to solve this to add a mechanism to queue a token
> callback.  The callback would be called during token posession after the
> current pending messages are transmitted.  Then remaining recovery
> efforts could be done when the callback is called.  If the recovery is
> still not complete, the token callback could be requeued for the
> remaining work until all work is completed.  
> 
> This would allow recovery to be done on every rotation of the token,
> until recovery is complete.  I think this solves this particular
> problem.  Does that mechanism work for the event service and AMF?  Would
> you be able to keep track of the state to synchronize in such a fashion
> as to work on the syncronization multiple times?
> 
> I can't think of any other solutions.

I have been thinking about how to do this.  So far, I had considered
using the plug support and sending one message at a time.  So, when I
needed to mcast a lot of messages for a recovery, I'd send them out one
at a time, when I received my own message, I'd send the next.  Once all
the messages had been sent, I'd clear the plug.  I haven't had a chance
to test this out yet (I see that you just got the plug support checked
in).  Anyway, I suppose that I could do something similar with the token
callback too.

Mark.


> 
> Thanks
> -steve
> 
> 
> ______________________________________________________________________
> _______________________________________________
> Openais mailing list
> Openais at lists.osdl.org
> http://lists.osdl.org/mailman/listinfo/openais
-- 
Mark Haverkamp <markh at osdl.org>