Next: ../SUPEUR93/scalability

Up: Presentation

Previous: ../STGIRONS/compare

The Situation:

Parallel scientific applications are typically

• written from scratch, or manually adapted from sequential programs

• using simple SPMD model

• in Fortran or C

• with explicit message passing primitives

Which means

• similar components are coded over and over again,

• codes are difficult to develop and maintain

• debugging particularly unpleasant...

• difficult to reuse components

• not really designed for long-term software solution