Data Analysis and Visualization B.K. Muite collaborators S. - PowerPoint PPT Presentation

Data Analysis and Visualization B.K. Muite collaborators S. Arshad, S. Aseeri, D. Acevedo-Feliz, O. Batra˘ sev, A. Bauer, D. DeMarle, M. Icardi, B. Leu, N. Li, A. Liu, E. M¨ uller, B. Palen, M. Quell, H. Servat, P . Sheth, R. Speck, M. Van Moer, J. Vienne, H. Yi benson.muite@ut.ee http://kodu.ut.ee/˜benson 16 April 2015

Outline • Motivation • Data from simulations • Challenges • Possible solutions

Motivation • Visualization examples

Volume rendering • https://www.flickr.com/photos/kitware/2864720427/in/pool-paraview • Electronic structure of a terpyridine molecule

Volume rendering • https://www.flickr.com/photos/kitware/2864720427/in/pool-paraview • Cross wind fire simulation

Iso Surfaces • http://www.paraview.org/Wiki/ParaView_In_Action#Magnetic_Reconnection_in_Earth. E2.80.99s_Magnetosphere • Magnetic reconnection simulation

Iso surfaces • https://www.flickr.com/photos/kitware/2864720427/in/pool-paraview • Convection simulation

Domain Partitioning • https://www.flickr.com/photos/kitware/2293739197/in/pool-paraview/ • Surface flow visualization by Renato N. Elias, Rio de Janeiro, Brazil

Volume Rendering • https://www.flickr.com/photos/kitware/2383791290/in/pool-paraview • Asteroid colliding with a planet. Simulation done using old version of Chombo.

Flow Field Visualization • https://www.flickr.com/photos/kitware/2293740417/in/pool-paraview/ • Visualization around Formula 1 Race Car by Renato N. Elias, Rio de Janeiro, Brazil

Flow Field Visualization • https://www.flickr.com/photos/kitware/2294528826/in/pool-paraview/ • Visualization around Formula 1 Race Car by Renato N. Elias, Rio de Janeiro, Brazil

Weather and Climate Prediction • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-09 • High spatial resolution, long time simulations

Computational Fluid Dynamics • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-07 • Optimize design of cars, trains, airplanes, ships

Computational Fluid Dynamics • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-08 • Optimize design of cars, trains, airplanes, ships

Computational Fluid Dynamics • http://en.wikipidea.org/wiki/File:JRC_N700_series_z28.jpg • Optimize design of cars, trains, airplanes, ships

Computational Solid Mechanics • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-24 • Optimize design of structures, consumer products, roads, bridges, cars, trains, airplanes, ships

Computational Materials Science • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-39 • Predict and control micro structural morphology to determine macroscopic characteristics, such as fracture resistance

Fusion • https://wci.llnl.gov/simulation/computer-codes/visit/gallery/gallery-38 • Predict and control instabilities

Time for IO 3 10 Computation Time 2 10 1 With output 10 No output Ideal 0 10 0 2 4 10 10 10 Number of Cores • Numerical solution of the Klein Gordon equation on Kraken (a retired Cray supercomputer formerly at the National Institute for Computational Sciences). A Fourier spectral discretization with 512 3 modes is used.

Time for IO 10 3 No output Images Output Ideal 10 2 Computation Time (s) 10 1 10 0 10 1 10 2 10 3 Number of Cores • Numerical solution of the Klein Gordon equation on Beacon. A Fourier spectral discretization with 512 3 modes is used.

Time for IO Scaling on VSC2 10 3 COPROCESSING ideal NO OUTPUT ideal REGULAR OUTPUT ideal 10 2 time in s 10 1 10 0 10 1 10 2 10 3 10 4 Number of cores • Numerical solution of the Klein Gordon equation on Vienna Scientific Cluster 2. A Fourier spectral discretization with 512 3 modes is used. Results obtained by M. Quell

Time for IO • Visualization workflow on K computer. A. Ogasa, H. Maesaka, K. Sakamoto, S. Otagiri “Visualization Technology for the K computer” Fujitsu Sci. Tech. J. Vol. 48 No. 3 pp. 348-356 (July 2012)

Time for IO • Visualization workflow on K computer. A. Ogasa, H. Maesaka, K. Sakamoto, S. Otagiri “Visualization Technology for the K computer” Fujitsu Sci. Tech. J. Vol. 48 No. 3 pp. 348-356 (July 2012)

Lattice Boltzmann - Visualization • http://optilb.com/openlb/ • Paraview Demo

Challenges • Large data sets, 10 , 000 3 points on large supercomputer • Computation time can be at a premium • Do not want to do input and output • Want to do insitu visualization • Want to be able to find areas of interest automatically - time for human interaction can be slow.

Challenges • Many scientists learn programming because they need to, not because they want to • Scientific data visualization has generally required a factor of 10-100 less computational resources • For moderate simulations on a remote cluster, one can move the data to a local workstation • For large data sets, this becomes infeasible and many centers have a separate visualization cluster • For peta and exascale machines, it is not always possible to have a separate visualization cluster

Challenges • Changes in parallel computer architectures • Many are now heterogeneous, CPU + accelerator • Many accelerators are GPUs or have architecture that suits GPU programming models • A good opportunity for computer graphics programmers • For peta and exascale machines, it is not always possible to have a separate visualization cluster

VTK • Visualization Toolkit • Open source library of graphics primitives • Community involvement • Parallel support • Aiming to also have GPU support • http://www.vtk.org • Used on its own and in other visualization tools such as ParaView and VisIt

DAX toolkit • Visualization toolkit to give multithreaded parallelizm a high level abstraction • Main concept is to have worklets that are independent small codes acting on local data that can be scheduled in multicore environment • Open source library of graphics primitives • Community involvement • CUDA, OpenMP and Intel TBB backends • Rendering support through OpenGL • Uses finite differences to compute gradients • Marching cubes algorithm to calculate isosurfaces is implemented • http://www.daxtoolkit.org/ • Used on its own and in other visualization tools such as ParaView and VisIt

PISTON • Visualization toolkit to give multithreaded parallelizm a high level abstraction • Open source library of data parallel graphics primitives • Community involvement • Uses NVIDIA’s thrust library • CUDA and OpenMP backends • Some OpenCL support • https://datascience.lanl.gov/PISTON.html • Used on its own and in other visualization tools such as ParaView and VisIt

CINEMA • Insitu analysis tool • Open source tool • Community involvement • Create a reduced size data set at runtime which can then be explored in post processing • Use database of images collected from multiple angles to allow for interactive exploration at runtime • Can one use artificial intelligence methods? • https://datascience.lanl.gov/Cinema.html • Use other visualization tools such as ParaView, then links them together with a carefully constructed database

Open Areas • Programming models: OpenCL? CUDA? OpenGL? C? C++? MPI? PGAS? • How generate graphics primitive libraries that will run on a variety of architectures? • How show uncertainty in scientific visualization? • Is separation of domains still feasible? Will scientists need to learn about computer graphics?

Acknowledgments • KAUST Visualization laboratory • XSEDE Extended Collaborative Support Service • Kraken at the National Institute for Computational Sciences through the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number ACI-1053575 • Vienna Scientific Cluster 2 • KAUST Supercomputing Laboratory • The Beacon Project at the University of Tennessee supported by the National Science Foundation under Grant Number 1137097; • Kitware

References • Moreland “A Pervasive Parallel Framework for Visualization: Final Report for FWP 10-014707.” Tech Report SAND 2014-0047, Sandia National Laboratories, January 2014. • https://en.wikibooks.org/wiki/Parallel_ Spectral_Numerical_Methods • Bethel, Childs and Hansen “High performance visualization” CRC Press (2010) • Data Science https://datascience.lanl.gov/ • Dorier, Sisneros, Petreka, Antoniu, Semeraro “A nonintrusive, adaptable and user-friendly insitu visualization framework” HAL-INRIA pre-print 00831265 http://hal.inria.fr/hal-00831265

References • Fabian, Moreland, Thompson, Bauer, Marion, Geveci, Rasquin, Jansen, “The ParaView coprocessing library: a scalable, general purpose in-situ visualization library.” LDAV 2011 IEEE Symposium on Large-Scale Data Analysis and Visualization, (Oct. 2011), 89-96. DOI= http: //doi.acm.org/10.1109/LDAV.2011.6092322 . • Ka˘ ceniauskas, Pacevi˘ c, Bugajev, “Efficient visualization by using ParaView software on Balticgrid” Information Technology and Control, 39(2):108-115 (2010) • Lo, Sewell, Aherns, “PISTON: A Portable Cross-Platform Framework for Data-Parallel Visualization Operators” EGPGV pp. 11-20 (2012).