Today marks the beginning of the Supercomputing 2014 exhibition. From 8:30 am to 12:00 pm in room 394, the tutorial “In Situ Data Analysis and Visualization with ParaView Catalyst” by David H. Rogers (LANL), Nathan Fabian (Sandia), Thierry Carrard (French Alternative Energies and Atomic Energy Commission), and Andy Bauer (Kitware) will be presented. For additional insights into ParaView Catalyst, we interviewed Andy Bauer, an R&D engineer at Kitware, who will be presenting the tutorial.
Tell us a little bit about ParaView Catalyst. How and why did the project get started?
The project started about 6 years ago through an Army SBIR award. The goal of the project is to enable analysis and visualization during a simulation (often called in situ, co-processing, or coviz) instead of afterwards (i.e., post-processing).
What is the main objective of the toolset?
As I mentioned, the goal of the project is to enable in situ analysis and visualization. Although this is not a new concept, the Catalyst toolset provides a general and accessible way to enable in situ analysis and visualization. With the continued growth of high performance computing (HPC) systems, it is going to be important to maintain the ability to gain insight into simulation output. This will get harder as the processing power of these systems will continue to outpace file I/O. In situ processing is a way to avoid much of the file I/O while also using appropriate computing power to process desired information.
How would someone new to the Catalyst project start working with it?
There is a variety of information available on working with ParaView Catalyst, including a tutorial titled “In Situ Data Analysis and Visualization with ParaView Catalyst” at SC14 in New Orleans. Another great source is the ParaView Catalyst User’s Guide. To add to that, there is a good set of examples at https://github.com/Kitware/ParaViewCatalystExampleCode. The ParaView mailing list (firstname.lastname@example.org) is also a great resource for individual issues that come up.
What are the possible impacts of Catalyst? To what fields can it be applied?
One of the biggest impacts of Catalyst is probably the fact that users can utilize it to get more useful information out of their simulations. For a while, it has been too costly to write out all computed information from these large scale numerical simulations. So, in essence, analysts have been seeing only part of the results of their simulations, and they may have missed key aspects due to the fidelity at which they analyzed their results. By using Catalyst, they can access all of the simulation results and choose what information is important enough to keep. As to the fields to which Catalyst can be applied, the idea is that any information that ParaView can be used to post-process, Catalyst should be able to co-process.
How does Catalyst leverage ParaView?
There is a lot of functionality in ParaView, and Catalyst can access nearly all of it. For in situ processing though, a better way to look at it is to ask: What parts of ParaView can be excluded? The reason that this is important is that Catalyst will take up computing resources (e.g., memory, CPU cycles) during a simulation run, and we want this overhead to be as low as possible. By linking to all of the functionality in ParaView, a simulation code will also be saddled with storing all of ParaView’s, and thus VTK’s, libraries in memory.
If you are only using a small portion of ParaView, and most people are, then the unused parts are essentially a drain on valuable system resources. The Catalyst editions are a way to avoid this issue. These editions are lightweight versions of ParaView that have only the needed parts that are required to output information from Catalyst. We have a set of pre-configured editions that will suit many users’ needs. (See our previously posted blog entry for more information.) If these pre-configured editions are not appropriate, it is not overly complex to create a specialized edition to add in functionality to get exactly what is needed.
How will you be showcasing Catalyst at Supercomputing 2014?
There will be a variety of events showcasing Catalyst at SC14. I already mentioned the “In Situ Data Analysis and Visualization with ParaView Catalyst” tutorial. Nathan Fabian also presented on “Instruction Memory Overhead of In Situ Visualization Libraries on HPC Machines” on Sunday. During the Big Data Analytics: Challenges and Opportunities workshop, there will be a presentation titled “An Image-Based Approach to Extreme Scale In Situ Visualization and Analysis” by Jim Ahrens from Los Alamos National Laboratory. Additionally, there is a video in the Visualization & Data Analytics Showcase on ParaView Catalyst being used with the MPAS-Ocean climate simulation code, as well as several Kitware booth presentations.