RE: [Algorithms] Re: Zen of Networked Physics

[Jon W. <hp...@mi...>]

> As far as I can see, resending the "important moves" from the client side
> should just be done if an acknowledgment mechanism detects previously the
> packet loss.

The problem is that you then missed the deadline to transmit that 
change to other people in time to not make them snap. The idea behind 
"important moves" is that you re-transmit changes in each packet, 
until you have the change acknowledged. The side benefit of this is 
that you can delta compress the important moves when transmitting, so 
they really don't take much extra space at all. The other benefit of 
that system is that you can do the Quake III thing, and use the last 
acknowledged packet as baseline for future transmissions (although 
this is orthogonal to important moves).

> Is the standard way in wich a sliding window protocol such as TCP detects
> packets losses, and then (just then) resends the lost data.

That's exactly why TCP is so bad for real-time interactive experiences.

Cheers,

			/ h+