Game Development Reference
In-Depth Information
Figure 11.5 The objects located outside the frustum of the main camera are eliminated by the
application and are not sent to the graphic card
criteria and then these machines display the data. This method is a little more complex
to implement as it requires developing a wrapper to intercept the calls made to this
graphic card. Programming a wrapper can be as “simple'' as replacing the software
library (VRGL) or can even require a complete software package to be developed
(Syzygy, More3D, Tech-Viz). Another disadvantage of this approach is that the entire
3D database needs to pass through the network for each frame, which requires using a
network specially meant for the distribution of rendering. But the biggest advantage of
this method is its universality: The source code of an application is not required to be
able to apply this technique, as the wrapper intercepts the application's output data.
This second approach generally looks more attractive because, at first sight, it
seems possible to distribute the rendering of any application on multiple machines.
This universality is, however, very relative. In fact, if the application is not initially
meant for distribution of rendering, we can stumble upon a problem that at first seems
irresolvable. The 3D applications are in fact often optimised to send the graphic card
nothing but the objects (triangles) that are likely to be displayed, i.e. the objects in the
observation zone (frustum) of the camera (frustum culling technique). The advantage
of this technique is that the graphic card is not overloaded with objects that will not be
visible anyway! But in our case, this means that the 3D objects which will not be visible
from the point of view of the main machine will be simply eliminated and thus not sent
to the graphic card (see figure 11.5). As a result, it will be impossible to intercept these
objects to redirect them to another machine to display them. Even though it is possible
to restrict its effect by defining a large focal length of camera in the application, this
problem certainly is the biggest limitation of the call interception method.
Distributable parameters
Usually, it is possible to distribute the calculations of stereoscopy (one machine com-
putes the image meant for the right eye while another machine does the same for the
left eye) and/or to split the total image into multiple zones and assign one machine for
Search Nedrilad ::

Custom Search