A rowwise block-striped partitioning strategy (Figure 4)
is used to divide the landscape across the 32 processors.
With this data partitioning method, the landscape is divided into
groups of complete contiguous rows [KGGK94]. Due to the large
computational demand
of both the hydrology and vegetation components
and the irregular shape of the study area,
each processor is assigned a group based on the total number of
500m grid cells (Figure 5), as opposed to the
total number of rows, in order to
achieve a similar initial workload balance. This partitioning method
simplifies the deer movement process (to be discussed in
Section 3.3.5), since each processor has only two nearest
neighbors. Processor PN
's nearest neighbors are
PN
and PN
, with the exception of PN
, which has
no nearest neighbor to the north, and PN
, which has no nearest
neighbor to the south.
Figure 4: Example rowwise block-striped data partitioning with 16 rows
and 4 processors.
