In-Situ PBVR
(In-Situ Particle Based Volume Rendering)
ML-PBVR v1.0 (2024/03/01)
ML-PBVR ver. 1.0, a multi-point remote visualization application, is now available for download. ML-PBVR allows you to visualize a large-scale simulation on a remote supercomputer on your PC at the same time as it is computed. ML-PBVR enables VR visualization using a head-mounted display (HMD), and allows interactive analysis among users by allowing multiple users to change visualization parameters.
The source code is available from github.
https://github.com/CCSEPBVR/CS-IS-PBVR/tree/release_multi_location_pbvr_v1.0.0
Download
IS-PBVR v2.2 (2023/11/10)
The latest version of PBVR2.2 is available for download. This update enables CS/IS/VR-PBVR to visualize textured polygon data (FBX/3DS format) and 3D point cloud data (LAS format). In addition, the source code for the CS/IS/VR-PBVR client program has been integrated.
The source code is available from github.
https://github.com/CCSEPBVR/CS-IS-PBVR
Download
Binary |
|---|
IS-PBVR 2.1 (March 27, 2023）
IS-PBVR v.2.1 is now available for download. This update increases the number of element types that can be processed in IS-PBVR's unstructured grid. In addition, the Mac ARM architecture is now supported.
The source codes are available from release_is_pbvr_v2.1.0 branch in github.
https://github.com/CCSEPBVR/CS-IS-PBVR
Download
Binary |
|---|
IS-PBVR 2.0 (December 12, 2022）
The latest version of IS-PBVR 2.0 is available. In this update, source codes of CS/IS-PBVR’s client are integrated, and pre-defined colormaps are updated.
The source codes are available from github.
https://github.com/CCSEPBVR/CS-IS-PBVR
Download
Binary |
|---|
In-Situ PBVR v1.2.1 (July 27, 2022）
The latest version of In-Situ PBVR v1.2 is available.
In this update, the composition of the volume rendering of the simulation result and the boundary shape polygon becomes possible.
In-Situ PBVR v.1.2.1 is available, which is bug fixed version of v.1.2.
Download
Binary |
|---|
Source code |
|---|
Old Version
Purpose and outline of program development
As the performance of supercomputers has improved, computing power has increased from tera-scale to peta-scale to exa-scale. Visualization is an essential technology for gaining scientific knowledge from the data generated in simulations. However applying conventional visualization strategies for exa-scale data is difficult with current infrastructure, as network speeds are limited for transferring data to client PCs, as is the RAM and storage available on the clients.
For peta-scale simulations, a client / server visualization protocol is emphasized, where data generated on servers with abundant computing resources are transferred to clients for visualization. However there is concern that data output to client storage itself will be a major bottleneck for exa-scale computing.
In-situ visualization, whereby images are generated on the same environment as the simulations, is a method that is expected to provide efficient visualization for exa-scale simulations. However, conventional in-situ visualization has the following problems.
１．Visualization failures often occur because visualization parameters such as viewpoint position, color and opacity are set before batch processing.
２．The cost of visualization processes suppresses simulation performance as domain decomposition for stencil calculations causes data reconstruction, or because global communications are required.
３.Techniques for visualizing multivariate data from multiphysics simulations are insufficient.
The interactive In-Situ Visualization Framework (In-Situ PBVR), based on the Particle-Based Volume Rendering (PBVR) technique, solves the above problems and can be used to visualize large-scale calculation results obtained from exa-scale simulations.
１．In-Situ PBVR compresses large volume simulation data in to small visualization particle data in the in-situ environment and then transfers the particle data to the client PC for remote interactive visualization. Interactive changes of visualization parameters are made possible by exchanging data between the client and server via a daemon process.
２．The visualization process can be parallelized without changing the domain decomposition, and high scalability can be achieved because it does not require expensive inter-node communication.
３．In-Situ PBVR implements a multivariate visualization function that can be used for various simulations. The function enables composition of volume data and composition of color/opacity functions using user specified algebraic expressions. The computational algebraic system enables multivariate visualization during the simulation runtime.
Program features
・The framework consists of a particle sampler coupled to the simulation code, a daemon that controls interactions by processing files on storage, and a PBVR client that renders visualization results and allows interactive editing on the client PC.
・In addition to a hybrid MPI / OpenMP programming model, the code supports SIMD operations used in the latest many core architectures.
・By inserting a particle generation function into the simulation code, the particle sampler performs visualization processing simultaneously with the batch process of the simulation.
・An ssh port forward connection between the daemon and the PBVR client enables interactive visualization of simulation results during batch process execution.
・Structured grids, unstructured grids (hexahedron) and hierarchical grids (Block Structured AMR) are supported as simulation grids.
・An advanced transfer function design visualizes not only volume rendering but isosurface, cross section and multivariate data.
≪In-Situ PBVR framework≫
≪Synthesis of volume data and transfer function for multivariate analysis≫
≪Calculation performance on the supercomputer Oakforest-Packs equipped with Xeon Phi7250≫
A strong scaling test of In-situ PBVR was performed connected to a thermal flow analysis code of about 100 million grids.
≪Comparison of overall performance with traditional visualization apps≫
Network bandwidth is ～11 MB / sec. Visualization was 30 times faster with In-Situ PBVR.
In-Situ PBVR |
ParaView |
|
|---|---|---|
Data size [MB] |
581 |
5144 |
Particle generation [sec/step] |
38 |
- |
Transfer [sec/step] |
52 |
456 |
Image production [sec/step] |
7.7 |
900 |
≪Visualization results of heat flow analysis.（Left：In-Situ PBVR、Right：ParaView）≫
In-Situ PBVR gave equivalent results to the conventional visualization app.
Reference
[1] T. Kawamura, T. Noda, Y. Idomura, “Performance Evaluation of Runtime Data Exploration Framework based on In-Situ Particle Based Volume Rendering”, Journal of Supercomputing Frontiers and Innovations, Vol 4, No. 3, pp. 43-54, 2017.
[2] T. Kawamura, Y. Idomura, H. Miyamura, H. Takemiya, “Algebraic Design of Multi-dimensional Transfer Function Using Transfer Function Synthesizer,” Journal of Visualization, Vol.20, No. 1, pp.151-162, DOI 10.1007/s12650-016-0387-1, 2016.
[3] T. Kawamura, T. Noda, Y. Idomura, “In-Situ Visual Exploration of Multivariate Volume Data based on Particle Based Volume Rendering,” Proceedings of ISAV2016 in SC16, ACM Digital Library and IEEE Xplore, pp.18-22, DOI 10.1109/ISAV.2016.9, 2016.
Developer
Computer Science Research and Development Office, Center for Computational Science & e-Systems, Japan Atomic Energy Agency
Contact
ccse-quad(at)ml.jaea.go.jp
※ Substitute @ for (at).