As problems grow computationally, requiring much more memory resources or time to solve them, we must turn to using multiple processors to solve them. No only are we then able to solve larger and larger problems but we can also take advantage of the multiple processors to solve existing problems more quickly.
For our research, we have used the Charm++ FEM framework in order to parallelize our code. This framework is ideal for standard CVFE codes and therefore has been easily integrated into our dynamic version. All of our simulations are run on Pentium III, 600MHz processors running Mandrake Linux.
As a test of the parallelization we selected the L-angle problem using dynamic insertion at the stress level. The simulations were run for time steps on 1, 2, 4 and 6 processors. The dynamic insertion interval was fixed at 500 time steps. As expected, the solutions for each of the simulations were identical, although the speedups decreased when more processors were used as seen in Figure 4.47. This decrease in speedup is primarily due to the serial portion of the code. Since most codes are not fully parallel, rather they have some serial portions, the relative solution times for the serial and parallel portions of the codes grow closer as the number of processors increases. In our problem, the serial portion corresponds to the initial reading of input files as well as the periodic cohesive element insertions and mesh repartitioning occurring every 500 time steps. The speedups can be increased if the insertion interval is increased and if the pre-processing of input data can be improved.