Agent Architecture for Mesh Based Simulation Systems
The paper presents an analysis of requirements for building simulation systems with tightly coupled components, such as typical mesh based PDE approximation software. The considered systems are characterized as having high communication to computation ratio. When designing architectures for such systems the hardware and middleware capabilities for providing communication links between processes have to be investigated and fully exploited. This is the place where agent technology perfectly fits the requirements. In the whole system, the capabilities of agents should be complemented with less flexible but more efficient software organization.
As an example a framework for finite element simulations, employing a modular architecture (described in ), is considered. Communication requirements for typical computations are estimated and evaluated in view of possible inter-process communication. The role of agents in setting up the execution structure of simulations is described.
KeywordsDomain Decomposition Message Passing Hardware Resource Linear Solver Modular Architecture
- 2.Szymanski, B., Varela, C., Cummings, J., Napolitano, J.: Dynamically reconfigurable scientific computing on large-scale heterogeneous grids. In: Wyrzykowski, R., Dongarra, J., Paprzycki, M., Waśniewski, J. (eds.) PPAM 2004. LNCS, vol. 3019, pp. 419–430. Springer, Heidelberg (2004)CrossRefGoogle Scholar
- 6.Płażek, J., Banaś, K., Kitowski, J.: Comparison of message passing and shared memory implementations of the GMRES method on MIMD computers. Scientific Programming 9, 195–209 (2001)Google Scholar
- 7.Banaś, K.: The application of the adaptive finite element method in large scale computations (in Polish). Wydawnictwo Politechniki Krakowskiej, Kraków (2004)Google Scholar