Page 99 - Anatomy of a Robot

03_200256_CH03/Bergren 4/17/03 12:27 PM Page 84
84 CHAPTER THREE
The technique in the previous example cannot be extrapolated to a much more com-
plex computational problem. As the statements in complex programs grow in size, the
number of “branches” increases rapidly. In the previous If statement, only 1 branch was
1
used, so we only needed about 2 processors (3 actually). In more complex programs
with many branches, the number of processors needed grows very rapidly, making par-
allel processing impractical. The way to avoid this problem is to restrict the number of
applications we try to tackle with parallel processing.
Many classical computational problems can still be partitioned naturally into paral-
lel tasks. Consider weather processing or vision systems (for the robot). The field of
view can be partitioned into areas, and a single processor can be assigned to each area
in an array. Each processes the information coming into its area. Generally, the proces-
sors can communicate with their neighboring processors. In a weather application, each
processor updates the weather in its small area (which may only be a few hundred
meters square). It communicates with its neighboring computers to inform them about
relevant events, such as moist air moving into their area. In such a way, weather fore-
casts have been made much more accurate and timely. The array processor has the gen-
eral structure shown in Figure 3-5.
Such an array can be built using general-purpose processors, but companies have cre-
ated processors specifically designed for parallel processing. They contain communi-
cation structures and special instructions that make parallel processing more efficient.

4 Parallel Processors
Predict Weather

Data

4 US Weather Zones

FIGURE 3-5 Parallel processors can divide up calculations for weather
prediction.