[interfacekit] Re: BHandler::fToken

"Marc Flerackers" <mflerackers@xxxxxxxxxx> wrote:
> > When we are using BMessages (which are itself not expensive, I 
> > agree
> > with you here), we lose the possibility to bundle several messages
> > together and send them once. And there is the real speed 
> > bottleneck,
> > not in the creation of the message.
> But by caching/bundling messages it takes longer before they arrive, 
> so they
> can have lost their meaning already (mouse was here some time ago 
> :)), and
> the responives of the system may go down.

Sure, but that depends on how you do it :-)
Right now, messages are lost ways too often in the windowing context.

> > Also, there are some things (like BView pointers) that only the 
> > BWindow
> > knows about (at least, I think so), so it may reduce overhead there 
> > as
> > well (as creating & flattening on the server, copying, unflattening 
> > on
> > the client, and then changing, and passing it to the target looks 
> > quite
> > expensive compared to the app=3D5Fserver messages).
> Well, if we find out that the BMessages form a real bottleneck, we 
> can
> always try to optimize :). But I never found my BeOS system slow in 
> response
> to mouse movement. I think most of the optimization can be found in 
> using
> the most from the newer hardware. Especially graphics accelleration.

Er... I thought R5 would do the bundling, too=3F Does it really send 
BMessages directly=3F

> Has anyone tried BeOS R5 on a processor with Hyper-Threading enabled=3F 
> I was
> wondering how much of a performance boost you get.

I am not sure it'll work - at least, you shouldn't do it in a MP 
environment (it wouldn't be nice if the first two virtual processors do 
all the work (being CPU 1) while the other two do nothing (i.e. CPU 2 
sits idle)).

Adios...
   Axel.

Other related posts: