An Interactive Environment for Numerical and Combinatorial Supercomputing
Star-P is a flexible interactive system intended to enable computational scientists and engineers to use a high-level language to program high-performance parallel supercomputers and commodity clusters. The Star-P user writes code in the Matlab language. Parallelism is available via data-parallel operations on distributed objects, and via task-parallel operations on multiple objects. The semantics of the parallel data objects are implemented by operator overloading in Matlab; the underlying parallel implementations of the primitive operations are from various parallel software libraries. Star-P includes distributed sparse matrix data structures and operations, which allow rather simple and elegant descriptions of various combinatorial and graph-theoretic computations. In this talk I will introduce Star-P, and describe a toolbox for combinatorial scientific computing that we are currently building for it.
