On the development of MP-TOUGH2

The authors are developing MP-TOUGH2 for exploiting massively parallel computers. The goals of this effort are to (1) create a data-parallel subsurface transport code for solving larger problems than currently practical on workstations, (2) write portable code that can take advantage of scalability to run on machines with more processors, and (3) minimize the necessity for additional validation and verification of the resulting code. The initial strategy they have followed is to focus on optimizing the generic and time-consuming task of linear equation solution while leaving the bulk of TOUGH2 unmodified. In so doing, they have implemented a massively parallel direct solver (MPDS) that takes advantage of the banded structure of TOUGH2 Jacobian matrices. The authors have compared timings of the iterative conjugate gradient solvers DSLUBC, DSLUCS, and DSLUGM written in Fortran77 for the front end with the MPDS which uses the data parallel unit. The MPDS shows good performance relative to the iterative conjugate gradient solvers on the free-convection test problem. The robust direct solution provided by MPDS can be used to (1) check on the veracity of a given iterative conjugate gradient solution, or (2) be used on certain problems where iterative solvers fail to converge. The test problem used in this study is the Elder pure thermal convection problem. For generality, the authors use the TOUGH2 equation of state module EOS3 for water, air, and heat, although this particular problem is a single-phase flow problem