|
Seminar Synopsis
Future Challenges for Scientific Simulation
John R. Rice
Computer Sciences Department
Purdue University
West Lafayette, Indiana
Past and future trends for raw computing power and for software productivity were
the topics of Rice's seminar. These trends will greatly enlarge the scope of feasible
scientific simulations, said Rice. He discussed the scientific challenges to achieving
these simulations and presented potential approaches.
Available computing power (determined by hardware and algorithms) directly determines
the feasibility of potential approaches to simulation. In recent decades, hardware
power has grown at an astounding rate. Equally astounding advances in algorithm power
have been achieved. The combined effect of these advances is to increase our power
by 12 to 20 orders of magnitude for many real simulations.
Programming is another key component in simulation. Increases in the productivity
of writing computer codes in FORTRAN, C, Java, etc. are very low. The result is that
programming is now the principal cost in simulation, and therefore new approaches,
such as problem-solving environments, are needed. Five future challenges for scientific
simulation were discussed in some detail. These challenges are: multi-physics phenomena,
multi-scale phenomena, control, validation of simulations, and improved algorithms.
The first three are application areas that are moving into the feasible range because
of increasing computational power. Validation is increasingly important as the complexity
of the simulations increases. Not only does uncertainty increase with complexity,
but the penalty for failure also increases. The final challenge, said Rice, is that
increased hardware power is not enough-we must (and can) also increase algorithm
power. |
|
 
|
| |
|
|
