Computations on clusters and computational GRIDS encounter similar situations where the processors used have different speeds and local RAM. In order to have efficient computations with processors of different speeds and local RAM, load balancing is necessary. That is, faster processors are given more work or larger domains to compute than the slower processors so that all processors finish their work at the same time thus avoiding faster processors waiting for the slower processors to finish. In addition, the programming language must permit dynamic memory allocation so that the executable size is proportional to the size of the partitions. The present version of the AERO code uses the F77 programming language which does not have dynamic memory allocation thus the size of the executable is the same for all processors and leads to situations where the RAM for some processors is too small to run the executable. In this report, we extend the parallel F77 AERO code to F90 which has dynamic memory allocation. The F90 version of the AERO code is mesh independent and because memory is allocated at runtime and memory is only allocated for the code options actually used, the size of the F90 executable is much smaller than the F77 version; as a consequence many tests cases that cannot be run on clusters and computational GRIDS with the F77 version can be easily run with the F90 version. Numerical results for a mesh containing 252K vertices using 8-nina and 8-pf processors running on the MecaGRID using GLOBUS using GLOBUS and heterogeneous partitions resulted in a speedup of 1.45 relative to the same run using the homogeneous partitioning which compares well with the theoretical speedup of 1.5. This validates the efficiency of the F90 version of the AERO code.