In-Situ Visualization

After a break, we reconvened to talk about recent improvement in ParaView’s In-Situ support.  It’s based heavily on the already existing PVBatch tool but offers bindings where you can link a PVBatch interpreter directly into your code.  You instantiate a special reader as a singleton and pass in pointers to the data in-memory, and then you can execute pretty much any PVBatch python script to perform your analysis or visualization.

Live during the presentation he compiled and ran a simple in-situ visualization that simulated a few basic particles and rendered them as images on-disk.

The details are pretty involved and technical, so I highly suggest you take a look at his slides and review the ParaView documentation if you’re interested.

Statistics

The last talk involved some new statistics support they are integrated into ParaView.  The new filters include Contingency tables, Principal Component Analysis (PCA), k-means clustering, multicorrelative, descriptive statistics, and bivariate histograms.  He showed examples of several of the various operations, and a video of some of them is available on the ParaView Wiki.

Conclusions

The guys at KitWare are still going strong on ParaView development, and the users at the various labs using it on a daily basis are actively contributing to new features and functionality at a surprising rate.  After the talk, several people talked about some of the difficulties getting Paraview to work in the HPC environment, issues like the massive filesize, dependency on dynamic libraries, and client-server port restrictions.  KitWare is aware of the problems and actively pursuing solutions.  One that we should see soon is the ability to selectively enable and disable portions of ParaView, far beyond the current “With GUI/Without GUI” options currently available, so that you can compile a minimal version with only the filters you require, making it better-suited for in-situ usage.

Definitely want to stay up-to-date on future advancements.

Previous
Tags