HPC & In Situ

HPC

The ParaView team continues to be on the cutting edge of using high-performance computing to address your visualization needs. The team is an active participant in the Exascale Computing Project, a multi-laboratory collaborative initiative by the Department of Energy involving the Office of Science and the National Nuclear Security Administration. ParaView has been deployed on Frontier, the world’s fastest supercomputer* with 1.1 exaflops of performance.

As a result, ParaView scales to allow visualization and analysis of large scientific datasets using the aggregate disk space, processing power, and memory of a cluster or supercomputer. Its family of MPI-enabled executables runs in parallel on distributed memory computers, where each node in the machine processes only a small portion of the entire data. The many results are combined and rendered in parallel when the resulting geometry is large or in serial when the geometry is small. 

And with ParaView’s client/server architecture, you can perform your computations remotely on an cluster or supercomputer and view the results using your local desktop or laptop.

In Situ

I/O will always be a contributing factor in terms of overall performance on HPC machines.  Traditional simulation workflows that run a simulation and then post-process the results after the simulation completes require a large amount of relevant information for each time step to be saved on a disk. This can result in high amounts of relatively slow I/O compared with the information required to view the visualization results.

Reduce Your File Size with ParaView

Full data set – 448MB
Surface of blades – 2.8MB
Image – 71KB

The numbers above shows that if you could generate the visualization in situ instead of afterward, the amount of saved data could be drastically reduced. Less data can result in more time steps being rendered without dramatically impacting your computational time. The videos below show the same CTH simulation. The one on the left was done using the conventional post-processing approach, while the one on the right was done in situ. 

Catalyst

Catalyst2 is an in situ library with an adaptable application programming interface that enables easy integration of analysis or visualization tasks within simulation codes. Combining the scaling capabilities of VTK and ParaView makes Catalyst2 a robust solution and can be integrated using either C++ or Python.