the collisional kerma approximation as a basis for absorbed dose computation. In order to use
deterministic techniques in high energy photon dosimetry applications, some special treatment
for secondary electron transport is therefore required; this will be solved using the EDK-SN
method.
2.5 Parallel Decomposition
For solving the 60C0 model problem in a parallel environment, different domain
decomposition strategies were considered. In Table 2-1, we present the CPU times and memory
requirements for the PENTRAN calculations using different quadrature sets. Parallel
decomposition results for solution of the 60C0 problem model are also presented in Table 2-1
based on a constant number of meshes of 126,000 with P3 aniSotropy for a single energy group
and different angular quadrature orders. From the table data, it can be observed that increasing
the number of directions scales up running times to the processor workload. More work on each
machine lowers the relative impact of message passing synchronization and latency, relative to
additional computational work, as evidenced where eleven times added workload only required
9.3 times as much computational effort (referring to the S42 calculation). This is a useful
illustration that emphasizes that with parallel computation, added model fidelity may not actually
cost additional wall clock time to yield a more robust solution.
Table 2-1. Total number of directions, cumulative problem time** required for a single energy
group, P3 scattering, 126,000 fine mesh cells using 12 processors on a parallel cluster
~Angular Numb~er Parallel Decomposition Cumnulative
quadrature of Proc~essors Iterations Problem T~ime Time
(relative # (minutes) ratio*
workoad)Angle E~nergy Space
S12 (1.0) 12 4 1 3 6 3 1
S22 (3.2) 12 4 1 3 7 11 3.2
S32 (6.5) 12 4 1 3 7 22 7.3
S42 (11.0) 12 4 1 3 7 28 9.3
* Directions and time ratio are referred to the Sez quadrature set. ** Non dedicated running tune