Resumen
In situ processing - i.e., coupling visualization routines to a simulation code to generate images in real-time - is predicted to be the dominant form for visualization on upcoming supercomputers. Unfortunately, traditional in situ techniques are largely incongruent with exploratory visualization, which is an important activity to enable understanding of simulation data. In re- sponse, a new paradigm is emerging: data is transformed and massively reduced in situ and then the resulting form is explored post hoc. The fundamental tension in this approach is between the extent of the data reduction and the loss in integrity in the resulting data. However, new oppor- tunities, in terms of increased access to data, may blunt this tension and allow for both sufficient data reduction and also more accurate analysis. With this paper, we describe the trends behind "data exploration at the exascale" and also summarize some recent results that confirmed that this new paradigm can produce superior results compared to the traditional one.