Re: Performance of Gnome and Gnopernicus

[Bill Haneman <bill haneman sun com>]

Michael/All:

Note that the blocking while delivering events to clients is, in fact, a
feature and not a bug.  That is to say, if we don't block at that point
it can create more problem than it solves for assistive technologies. 
Better to block until the notifications have gone out.

This doesn't at all explain a two-second delay however.  My benchmarks
have indicated that on modest machines we can process a few thousand
AT-SPI calls per second.  The maximum latencies in this part of the
process are really quite small from the "user time" perspective.  I
think that the visible lags are coming from the way the APIs are being
used, etc. 

regards

Bill

On Wed, 2003-05-28 at 11:37, Michael Meeks wrote:
> Hi Adam / John,
> 
> On Tue, 2003-05-27 at 04:42, Adam Myrow wrote:
> > I would like to know this as well. I hate to say it, but as pleased as I
> > am that Gnopernicus does something, I have to say that Gnome strikes me as
> > a very bloated, slow desktop.
> 
> 	For what it's worth - the problem is in fact not Gnome, but the latency
> inherent in the event emission mechanism that is used; we can speed it
> up very substantially with relative ease;
> 
> >  It also depends on what speech driver you are using.
> 
> 	This is prolly an architectural issue in gnopernicus / gnome-speech.
> Someone needs to sit down with a debugging ORBit2 and
> ORBIT2_DEBUG=traces:timestamps - and see what's going on.
> 
> 	One of the problem is that per event, the application hangs until the
> registry has emitted the events to each AT (serially, blocking until
> each has finished processing the event).
> 
> 	Another thing would be to do asynchronous unrefs / more aggressive
> caching; can you try doing a tweak for me and seeing if it helps ?
> 
> 	If you hack at-spi/cspi/spi_main.c (cspi_object_unref): and just #ifdef
> out the body of the method - that may save a number of round-trips, at
> the cost of some leakage at the client; it'd be interesting if that
> makes a noticable difference.
> 
> 	Regards,
> 
> 		Michael.
> 
> -- 
>  michael ximian com  <><, Pseudo Engineer, itinerant idiot
> 
> _______________________________________________
> gnome-accessibility-list mailing list
> gnome-accessibility-list gnome org
> http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list