User Friendly High Productivity Computational Workflows Using the Vision/HPC Prototype
Abstract
HPCs (high-performance computers) utilize multiple (e.g. hundreds or thousands of) processors to compute very large problems quickly by distributing the computation across many processors in parallel. This liberates problem conceptualization from the memory/storage constraints of a single desktop workstation. Unfortunately, the complexity of programming HPCs is off-putting for new users. Furthermore, most DoD users work from a Windows PC so that learning Unix well enough to parallel program is itself an obstacle. What is needed is a workflow by which simplifies the programming task in a familiar environment while leveraging the computational power of HPCs. VISION is a freely available, Python-based, drag-and-drop visual programming environment that programming for drawing flowcharts that encapsulate the underlying programming complexity. This means that computations are strung together by dropping and connecting computational boxes on a canvas instead of writing source code files. This is important for productivity since productivity is dominated by the time spent programming versus the time spent analyzing results. As a Python-based package, it is possible to embed parallel computing features from the open source iPython package into VISION to enable both visual programming and parallel execution on remote HPCs. This paper discusses the prototype we built at SSC-SD for a visual parallel programming workflow based on VISION and iPython for parallel computing using a Linux cluster as a backend and a Windows XP workstation as the front-end.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2008
- Accession Number
- ADA503636
Entities
People
- J. H. Unpingco
Organizations
- Ohio Supercomputer Center