Database Reference
In-Depth Information
Figure 9.2
(See color insert following page 224.) Shown is a 36-domain
dataset. The domains have thick black lines and are colored red or green.
Mesh lines for the elements are also shown. To create the dataset sliced by
the transparent gray plane, only the red domains need to be processed. The
green domains can be eliminated before ever being read in.
recent work rectifies this shortcoming, as well as provides a solution lay-
ered atop the RFB protocol to capture and deliver image pixels produced by
hardware-accelerated, distributed memory-rendering infrastructure.
16
In both
cases, the rendering infrastructure, particularly the image capture and remote
delivery, is transparent to the visualization application.
9.2.2 How Metadata Can Enable Optimizations
In this section, we give an example that motivates how the presence of meta-
data
∗
can lead to extensive optimizations for the visualization tool. I/O is the
most expensive portion of a pipeline execution for almost every operation a
visualization tool performs. We can reduce I/O and processing load by read-
ing only the domains that are relevant to any given pipeline operation. This
performance gain propagates through the pipeline, since the domains not read
in do not have to be processed downstream.
Consider the example of slicing a three-dimensional dataset by a plane
(Figure 9.2). In this case, most of the domains will not intersect the plane—
loading and processing those domains that do not intersect the slice plane is
∗
We define
metadata
as data about the total dataset to be processed, whose size is small relative
to the total dataset itself.
