[Verse-dev] its me again with an up to date pusher

[Emil Brink]

On Mon, 25 Oct 2004 16:14:17 +0200
Thorsten Jordan <Thorsten_Jordan at gmx.de> wrote:

> Hello,
> 
> the current (attached) pusher code waits longer, so that less packets 
> are lost. however it takes very long to push larger data.
> (1minute for the bunny, on localhost!)

I've started looking into it a little bit. I had one silly compilation
problem, line 707 calls mymodel->get_name() but the model.h I have does
not contain such a method. Might just be a case of partially old code,
though.

Further, the delay *before* verse_callback_update() in line 751 (the
call to mysleep(20)) is ill-advised. What you want to do is give Verse
the CPU quickly, so don't delay before calling verse_callback_update().

With this delay commented out on my laptop (1 GHz P-III), I can upload
the ppctest2.3ds scene in roughly 3,5 seconds, and the output reads
like this:

[...]
Phase 3: filling geometry layers...
Queue has 1 calls pending.
Needed 0 iterations of resending calls.
Queue has 81 calls pending.
Needed 0 iterations of resending calls.
Queue has 290 calls pending.
Needed 0 iterations of resending calls.
Queue has 17 calls pending.
Needed 0 iterations of resending calls.
Queue has 50 calls pending.

There is always "0 iteations" needed, which I take as a good thing.

More later, perhaps. :)

In other news, Brecht made me realize that the bogus ownership
reporting is just a bug, not a protocol issue. The server is simply
confusing the second instance of the client with the first, intern-
ally. This should be fixable.

Regards,

/Emil