Technical Details on Computation

The computational environment used to perform the parallel runs consisted of two separate Linux based clusters. The first cluster, GRIG, consisted of 64 dual-core nodes of Intel EM64T processors. The second cluster, FRODO, consisted of 62 dual-core nodes of AMD Opteron processors. The parallelization utilized the MPICH version 1.2.7p1 MPI library in conjunction with GCC version 3.3.5. The firebreak simulations are independent of each other and therefore provide for an embarrassingly parallel implementation based upon individual evaluations required by the GA.

The GA, implemented utilizing the PGAPack Library version 1.1, was parameterized with a population of 20,000 randomly chosen firebreaks. Each generation, 100 firebreaks were replaced, and the algorithm was run for 500 generations. Mutations occurred for both orientation and location of the replaced firebreaks. Mutations of firebreak location occurred in either a local and global manner randomly chosen for each mutation. Local location mutation limited the mutation within a 50 cell radius both vertically and horizontally of the initial location while global location mutation allowed mutation across the entire landscape. The replaced firebreaks were allowed to undergo both crossover and mutation in the replacement process. The GA performs 70,000 firebreak evaluations for each fire start location. A complete enumeration would require 750,000 firebreak evaluations, and future work utilizing "more realistic" firebreaks could easily render a complete enumeration computationally unfeasible as the number of possible firebreak configurations increase. Utilizing 30 processors, the average GA evaluation of a fire start location takes 105 seconds to implement with linear scaling on the number of processors.