Star-P: High productivity parallel computing

Star-P ‡ is an interactive parallel scientific comput-ing environment. It aims to make parallel program-ming more accessible. Star-P borrows ideas from Matlab*P [3], but is a new development. Currently only a Matlab interface for Star-P is available, but it is not limited to being a parallel Matlab. It combines all four parallelMatlab approaches in one environment, as described in the parallel Matlab survey [2]: embarrassingly parallel, message pass-ing, backend support and compilation. It also in-tegrates several parallel numerical libraries into one single easy-to-use piece of software. The focus of Star-P is to improve user productiv-ity in parallel programming. We believe that Star-P can dramatically reduce the difficulty of program-ming parallel computers by reducing the time needed for development and debugging. To achieve productivity, it is important that the user interface is intuitive to the user. For example, a large class of scientific users are already familiar with the Matlab language. So it is beneficial to use Matlab as a parallel programming language. Ad-ditions to the language are minimal in keeping with the philosophy to avoid re-learning. Also, as a design goal, our system does not distinguish between serial data and parallel data. C = op(A,B) print(C) C will be the same whether A and B are distributed or not. This will allow the same piece of code to run sequentially (when all the arguments are serial) or in parallel (when at least one of the arguments is distributed)