Masha Sosonkina
(organizer),
Iowa State University, USA
email: masha@scl.ameslab.gov
Parallel computing platforms advance rapidly, both in speed and in size. However, often only a fraction of the peak hardware performance is achieved by high-performance scientific applications. The main reason is in the mismatch between the way parallel computation and communication are built into applications and the processor, memory, and interconnection architectures, which vary greatly. One way to cope with the changeability of hardware is to start creating applications able to adapt themselves ``on-the-fly''. In accomplishing this goal, the following general questions need to be answered: How does an application perceive the system changes? How are the adaptations initiated? What is the nature of application adaptations? The talks of the minisymposium attempt to answer these questions by providing either an application-centric or a system-centric viewpoint. Given a parallel computing platform, the application performance may be enhanced by, for example, modifying the underlining computational algorithm or by using ``external'' runtime tools which may aid in better load balancing or mapping of scientific applications to parallel processors.