The following work packages and objectives were proposed for this six
month dCSE project with the aim to increase DSTAR scalability and
- Implementation of a 2D domain decomposition algorithm
Objective: DSTAR should scale to at least 10,000 cores with
80% scaling efficiency.
- To convert the base data structures for the new domain decomposition.
- To update the core fluid solver and run a basic validation test.
- To update the remaining multi-physics modules.
- To benchmark the new DSTAR code.
- Improved I/O for capability size data sets
Objective: DSTAR should be able to efficiently handle datasets for cubic grids of at least size 500 and beyond.
- To refactor the checkpoint procedures and statistics procedures for handling large data sets
- using MPI-IO type of parallel I/O solution.
- To re-engineer the logging procedures.
- To port some of the common I/O routines back to the
2DECOMP&FFT library. But only if this is worthwhile.
- Code refactoring. Most of this will be done along with the code update in WP1.
The remaining routines, in particular `user routines' will be modernised using Fortran 95
as these are regularly updated by scientists who use DSTAR. Some legacy library routines may be kept
in F77 form unless any significant performance issue is identified.
Objective: All user routines of DSTAR should be modernised.
Objective: All user routines of DSTAR should be modernised and a brief user document provided for future DSTAR users.
Lucian Anton 2011-09-13