Parallelism across the systemEpODE is capable to detect the separability of the ODE system into independent subsystems (a "natural" parallelism). Unfortunately, the other possibilities for parallelism across system were not implemented until now in EpODE, but we hope that in the next version new features in this direction will be added.
If there are at least two subsystems reported in the problem properties panel, and if we use the option Solve -> Multiprocessor code -> System distribution, then EpODE will create a number of processes equal with the subsystem number. A such process works on his associated subsystem and must communicate only with the main process which collects the results and display them. The communications and the process creation procedure are based on PVM routines. In particular, in this version of EpODE, the computation processes are associated to an executable program which must receive from the EpODE main program his subproblem and the current numerical method; the approximate solution calculus is the same as in the case of the sequential computation mode (to the corresponding dimensions).
We can speak about efficiency of the parallel/distributed algorithm implementation when we have a large number of differential equations or a large integration interval (especially stiff systems). Load balancing is obtained when the subsystems are of the same dimension and difficulty.