Big Visualization at Los Alamos Workshop on Parallel Visualization and Graphics LAUR-03-7666 Allen McPherson Los Alamos National Laboratory October 19, 2003 Outline Overview of our mission and computing environment Compute and visualization facilities Software infrastructure Production visualization process Future direction and challenges LAUR-03-7666 • • • • • What we do… • Mission – We develop and apply science and technology to… • Ensure the safety and reliability of the US nuclear stockpile • Reduce the threat of weapons of mass destruction, proliferation, and terrorism • Solve national problems in defense, energy, environment, and infrastructure • Goal LAUR-03-7666 – Creating and integrating core competency for Science Based Prediction of complex systems linking experiment, simulation, and theory Computational Environment • Current machines – Q (LANL) 20TF – White (LLNL) 12TF – Red (SNL) 3TF • Next generation LAUR-03-7666 – Purple (LLNL) 100TF – Red Storm (SNL) 40TF – Large computational clusters for “capacity” computing – 1PF (LANL) Functionality Requirements • Interactive visualization and analysis of extremely large, time varying, datasets – Also, canned multi-display movies • Visualization of such datasets over large geographic distance, yet still enabling interactivity • Support multiple display modalities LAUR-03-7666 • Same software on all modalities Generic Physical Simulations • Simulation of a shallow water asteroid impact LAUR-03-7666 • Simulation of a RichtmyerMeshkov instability experiemnt Visualization Facilities Switched analog distribution Office displays Collaboratories Theater Immersive facilities LAUR-03-7666 • • • • • Switched Analog Distribution LAUR-03-7666 • Supported over campus-like environment • Multiple switches: 128x128 & 30x30 Office Visualization • Stereo CRT through analog switch – 60 offices • High-end PC with IBM, et al. highres LCD display GIG-E connected 25 offices EnSight client/server model Users would like stereo here too LAUR-03-7666 – – – – Collaboratories • Delivering visualization facilities directly into user office areas • No visualization expert required • Bright, stereo walls allow normal lighting – Two built in SCC LAUR-03-7666 • Two more non-stereo walls • Additional office-type equipment installed for drop-in use • All driven through switched analog with our rendering infrastructure Visualization Theater • • • • • • Completed April 2003 22’ x 12’ rear-projected power wall 24 bright, stereo projectors 31M pixels 85 seats Driven through switched analog with SGI – Will drive with cluster soon LAUR-03-7666 • Presentation, reviews, training • Brings visualization performance to major decision making meetings of the Laboratory • Used on daily basis for large and small groups Immersive Environments • RAVE – Reconfigurable CAVE – Full immersion (VR) environment – Head tracking LAUR-03-7666 • An experiment in methods for understanding data Next-Gen Immersive Environment Spring/Summer 2004 5 rear-projected surfaces – – – • • • 33 projectors, 43+ M pixels Bright, digital projectors This builds on the positive user experiences in the RAVE: – – • Front wall: 15’ wide x 12’ high Floor/Ceiling: 15’ wide x 8’ deep Side walls: 10’ wide x 12’ high Immersion helps people understand simulations more quickly. They help bridge the comprehension gap between simulation scientists and experimental scientists and engineers. This will be driven with a new SGI 3900 w/ Voyager graphics. – 34 commodity graphics pipes LAUR-03-7666 • • Rendering Infrastructure • 3 SGI Onyx 2000, each with: – 16 IR pipes (48 pipes total) – 128 PEs, 128 GB memory, 5+ TB disk – All available through analog switch • High-performance desktop systems are used as a capable rendering system for moderate-sized problems. • A new SGI Onyx 3000 w/ Voyager graphics is being deployed to support – additional offices – the next-generation CAVE system LANL’s first production, distributed-memory, cluster-rendering system is currently being deployed for the SCC Theater. – Next step is to provide distributed-memory cluster-rendering support for offices & PowerWalls. LAUR-03-7666 • Software Environment • CEI Ensight – Mainline production visualization tool • Kitware ParaView and VTK – Open source, research tool – Lets us use other lab’s work (e.g. Sandia Ice-T) – Being deployed for some user functions • Comparative visualization • Kong LAUR-03-7666 – Parallel distributed version of Simian volume renderer – Collaboration with University of Utah (Joe Kniss) Distributed EnSight Architecture GigE network ASCI Q Viz Nodes Switched Analog Fiber Visualization Rendering Server EnSight Server-ofServers EnSight Client EnSight Server • Leave data in place where computed – – • • • Don’t store it twice Don’t size end-node systems to platform sizes Uses modest amount of bandwidth Supports interactive analysis Supports various display modalities LAUR-03-7666 PFS SGI Onyx, PC Desktops, or soon: clusters How is Visualization Used? • One big example: milepost run on White – – – – – – Ran for couple of months 650TB of data written (restart files) 364 visualization time steps 85M to 468M unstructured cells per time step Multiple variables per cell 21TB total visualization data LAUR-03-7666 • Run at Livermore, visualized at Los Alamos Big Viz Example • Visualization is used to… – – – – – – Understand results and communicate among ourselves To compare results of previous runs To ultimately document and present what we did To debug the codes To aid with code setup To do various qualitative and quantitative analyses LAUR-03-7666 • These are very difficult problems we are trying to understand • Use large number of visualization and analysis techniques to answer these questions • Some of these are billion dollar questions • Visualization is essential (per the users!) Big Viz Example • For this run… – Users ran many 2D preparatory studies • Did visualization and analysis with EnSight – Users ran a small number of 3D tests • Did analysis and visualization with EnSight – Ran EnSight visualizations while big 3D code was running • Not computational steering • Monitor progress of code • Get ahead of the curve: problem space is well understood LAUR-03-7666 – Users spent 6+ months doing viz and analysis on 21TB Big Viz Example • Remote visualization to Los Alamos from White – Obviously can not transfer the data from LLNL • Or even from Q to the LANL viz machines for that matter – OC-48 connection – 50 EnSight servers running on White • Read and extract geometric data (e.g. iso-surface) – Geometry sent to LANL and visualized on SGI pipes – Geometry extraction results in about 5% of the data (avg.) • So, for ~500M cells, can get 50M triangles (2-1 ratio) – This works extremely well LAUR-03-7666 • Users would like it faster though What’s Next • Visualization has been extremely successful at LANL – Has really enabled codes to advance at rapid rate – Users are pleased with production results – Visualization presentations really help executives to understand what we’re doing • And this in turn helps the Laboratory • But, users are demanding more and better services – More rendering resources – More offices served – Faster and more useful on large data LAUR-03-7666 • Research teams continue to work to transition results into production What’s Next: Clusters • Commodity cluster rendering – All DoE labs have been working on this for years – Now time to transition to production use • Initial effort underway at Los Alamos – 64 node cluster to be use exclusively to drive theater wall – 24 nVidia 3000G cards, 40 nVidia 5900 cards • Stereo and genlocked – Point-to-point from cluster to projector using DVI modems – CEI developing cluster version EnSight client • Same client/server paradigm: can viz remote data on cluster! • Using Chromium LAUR-03-7666 – Research tools will run on cluster What’s Next: Digital Video • We need to serve more offices with all this additional rendering power – We will more than double our existing rendering resources with the 64-node cluster – We will be adding many more clusters over the next few years (64-512 nodes) – Need to devise a way to feed offices, collabs, theaters, etc. • Large DVI switches not available LAUR-03-7666 – Lots of work going on in this area, some funded by DOE – At low costs it may be feasible to dedicate cluster nodes to offices or displays and simply go point-to-point over DVI modems Questions? • Thanks to… LAUR-03-7666 – Bob Kares: chief Los Alamos visualization user – Bob Tomlinson: Los Alamos VIEWS program manager