Lead Investigators : Drs. Gelb, van Dommelen
Code implementation is one of the major obstacles to solving cutting-edge large-scale problems in computational mechanics. While much of it is not sophisticated, there is a considerable effort involved in order to effectively utilize advanced computational resources at their true capabilities. It is difficult for an individual researcher to invest the required time. An interdisciplinary center will allow FSU researchers to share the load and thus become a leading force in state-of-the-art computational mechanics.
For efficient implementation, a researcher is required to become familiar with general parallelization ideas, specific coding techniques, as well as specific machine idiosyncrasies. Since hardware and software change rapidly, this requires a sustained and significant effort. When new students join, they too have to learn the required techniques. Computer codes written for specific machines also tend to become obsolete quickly when machines change, requiring laborious rewrites.
The interdisciplinary center in computational mechanics will be an enabling forum to allow FSU faculty to attack such problems. Fortuitously, it so happens that very few of the code-implementation details are specific to a particular area of computational mechanics. Through cooperation in an interdisciplinary center, faculty, postdoctoral researchers, and especially students can learn from each other. Researchers can share what works and what does not work, and they can share their knowledge about new capabilities, available software, buggy hardware and software, etcetera, potentially saving many months of laborious trial and error. Topical seminars, training sessions, and tutorials will be arranged to disseminate the garnered knowledge.
Software to take advantage of advanced computers such as PVM, MPI, and OpenMP is independent of the application area, and its knowledge can directly be shared. So can the implementation issues on whatever is the latest FSU computational hardware. Load balancing and communication issues will vary for individual problems, but the general ideas are common. Regardless of application area, large-scale computations are associated with large, unwieldy data sets that must be analyzed in order to extract the sought answers. Analyzing such data on the machine that created it, or compressing and transferring it to other postprocessing hardware are likely to be difficulties that are largely common between the areas. By sharing knowledge about such practical issues, the center will enable the efficiency of operation that makes addressing large-scale computational problems a reality.