Parallel simulation of ecological structured communities: Computational needs, hardware capabilities, and nonlinear applications

Previously, we developed a general scheme for parallel simulation of nonlinear structured population models where the critical development then was an efficient method of balancing the load among the available processors. We introduced the effects and necessity of load balancing together with the design of the load balancing algorithm for structured population models. Through the combination of parallelization by cohort, frequent rebalancing, and global birth combining we were able to demonstrate good scaling and speedup on 32-, 64-, 128-, and 256-processor machines. Here we return to the problem of parallel simulation for nonlinear structured population and predator-prey models and reconsider the need for parallelization given the improvement in computation hardware and software. Our conclusion is that even though there has been an extensive speedup in computers, the basic problem of managing the interplay of workload remains a fundamental problem for any parallel effort in simulation studies.