DescriptionParallel Computing: from Digital Twins to Grids to Deep Learning
We describe a community and personal journey in parallel computing starting (1980-1995) with simulations (perhaps now called digital twins) enabled to go from an at best illustrative two-dimensional model to the realistic three-dimensional case. We had computational science and the third paradigm of scientific study.
Then came the World Wide Web and the programming on it leading the Grid computing initiative (1995-2010). I still remember a caustic referee report from those early days that HTTP was too slow to be serious. We learned in the OGF (Open Grid Forum) how to build communities and eScience was a key theme. Distributed computing was king and pleasing parallel was coined.
Then came a new revolution (2010-?) with Big Data and the fourth paradigm of computing built on data science as the academic underpinning. Distributed computing remained but the Grid metamorphosed as data processing in the edge and fog. Parallel computing still continues in exascale scientific discovery but also in parallel deep learning.
We muse about this progression noting some common themes -- natural high-level parallel programming remains hard, an area where I had the pleasure of joint research with Ken Kennedy -- and messaging is critical in many different scenarios. We look to a spectacular innovative future where deep learning. simulations and parallel computing synergistically run together; each enhances the other to orders of magnitude better capabilities and performance.