Hypercube algorithms and implementations
SIAM Journal on Scientific and Statistical Computing
Supercomputers: algorithms, architectures, and scientific computation
C3P Proceedings of the third conference on Hypercube concurrent computers and applications: Architecture, software, computer systems, and general issues - Volume 1
ACM SIGAda Ada Letters
| Hi-index | 0.00 |
Emerging highly parallel multiprocessors offer an exciting alternative to conventional pipelined supercomputers for a variety of computationally intensive scientific applications. A factor that has impeded the introduction of these multiprocessor systems is that conventional languages, such as Fortran, cannot be directly used and new programming techniques must be mastered. A key issue for highly parallel systems is the development of appropriate programming environments.Programming environments are very difficult to specify for multiprocessor systems because a wide range of different programming paradigms are possible. For example, on a hypercube system, different applications require different operating systems for optimal performance. A typical hypercube programming environment includes a conventional language, such as Fortran or C, coupled with some message passing extensions. While the environment has a superficially familiar appearance, the organization of programs is much more complex and the user is now expected to program many functions which are conventionally done by the operating system. New programming environments need to be developed which have provisions for graceful fault tolerance, dynamic load balancing, dynamic algorithm selection, and automatic task decomposition and allocation.