The trend in HPC is that I/O rates are decreasing relative to the ability to generate data. In situ processing is increasingly being considered to deal with this trend. That said, in situ processing works best with a priori knowledge, since the desired visualizations can be arranged ahead of time. However, with exploration-oriented visualization, this a priori knowledge does not exist. As a result, a new model has been emerging that does both in situ processing and post hoc exploration. During the in situ phase, data is transformed and reduced. In the post hoc phase, this data is explored through visualization. The tension for this style of processing is between integrity and data reduction: too much data reduction can lead to very low integrity, while too high of a standard for integrity can prevent data reduction.
With this talk, I will describe recent research on using wavelet compression as the reduction operator for the in situ + post hoc paradigm. Specifically, I will discuss tradeoffs in integrity and data reduction, I will discuss whether the operator will fit within in situ constraints, and I will discuss how upcoming changes in architecture can be utilized to perform the operation even more efficiently.