Open Inventor and VolumeViz LDM Cluster Edition

Open Inventor and VolumeViz LDM Cluster Edition Powering oil & gas developers for next generation 3D scalable applications, from laptop and desktop to commodity clusters and advanced VR systems MERCURY WHITE PAPER By Daniel Lichau, Mike Heck and Thierry Dufour

DEVELOPER TOOLS FOR NEW CHALLENGES FROM LARGE DATA TO LARGE DISPLAYS Open Inventor from Mercury has become the de-facto standard for high-level 3D programming in the oil & gas industry. It provides robust support to major software vendors and corporate developers as well as research groups and innovative start-ups, across the full spectrum of visual applications including: Seismic data QC, pre-processing and interpretation Petrography, core analysis, borehole imaging Reservoir characterization, modeling Flow visualization, reservoir engineering Well planning, production management Well drilling, monitoring Engineering design, training and simulation The quest for petroleum resources presses software creators to meet requirements that appear increasingly harder to reconcile. How to improve human efficiency at all stages of workflows? How to manage larger and more complex data with higher accuracy to reveal more prospects and optimize exploitation? How to deliver faster to the market working solutions with cost effective deployment? VolumeViz LDM and Open Inventor Cluster Edition make it possible to solve this equation and rapidly develop robust, highly interactive, immersive or collaborative solutions that can manage 1 Gb 10 Gb 100 Gb 1 Tb Open Inventor and VolumeViz LDM Cluster Edition This powerful technology can dramatically improve the collaboration workflow in oil exploration and production, and significantly accelerate the decision-making process. POWERING YOUR SOLUTIONS In this paper we show how Mercury s 3D toolkits 1 can help you build solutions to meet your biggest challenges: Quickly implement user-friendly probing for seismic data with extended 2D slices or large 3D volumes, using VolumeViz. Enable roaming through huge volumes, whether on laptop or PC cluster, using VolumeViz Large Data Management (LDM). Take maximum advantage of the latest capabilities of graphics hardware for accurate, advanced rendering. Handle multiple data and attribute volumes, whether pre-computed or generated on-the-fly by CPU or GPU 2. Manage arbitrary display configurations including large multiscreen high resolution displays, for accurate analysis of detailed large scale data, enhanced group collaboration or presentations, taking full advantage of multi-pipe / multi-processor systems or PC clusters with Open Inventor Cluster Edition. Make your application VR aware. Simplify complex interpretation or drilling planning by easily turning a desktop application into a truly immersive VR application using stereoscopy, head tracking, a 3D interface and interaction. Avoid costly or unsafe data transfers and facilitate distant collaboration or resource sharing, by turning your application into a visualization server with remote rendering. Interact faster for probing and modeling in complex scenes with fast edit mode. Interactivity Other commercial solutions Simplify software development with high-level classes providing data representations for charts, meshes, flows, support for large models, high-resolution and vector hardcopy, and more. Easily integrate powerful 3D features using your existing environment, GUI toolkit, and language (C++, Java,.NET 3 ); then maintain your application with less effort. Leverage our professional support, expertise and services to build better solutions, more easily and more quickly. Data Size huge datasets even on commodity hardware architectures. Amazingly easy to integrate in applications, this toolkit enables your end-users to explore their largest data even on standard and easily extensible hardware, from laptop and remote lightweight clients to high-resolution display arrays. VOLUMEVIZ: PROBING INTO SEISMIC DATA 3D graphics has revolutionized the use of seismic data. Today, 3D visualization is pervasive in interpretation, modeling and planning, but it is also becoming crucial for acquisition quality control, real-time monitoring for safe drilling in complex reservoirs and other applications. 1 Open Inventor and VolumeViz LDM Cluster Edition 1 This paper covers features introduced in releases 5.04, 5.1 and 6.0 of Open Inventor, respectively released in June 2005 and planned for Q3 2005 and Q4 2005. Contact Mercury for more information. 2 Graphics Processing Unit 3 Open Inventor 6.0

Designed with input from our most demanding customers to allow rapid development of seismic visualization applications, the VolumeViz extension allows the inspection of 3D sampled data efficiently with a comprehensive feature set including: Mapping data on ortho/oblique slices, arbitrary surface geometry such as wells or horizons, volume skins, and of course direct volume rendering... Flexible probing by sub-volumes, 2nd level region of interest (exclusion box, fence, cross ROI ), clipping planes, clipping horizons/surfaces 4, powerful and customizable 3D interactors Mix multiple volumes and arbitrary geometry in the same scene. Predefined and application-defined transfer functions controlling data opacity and color to best highlight relevant data in real time. Selection/picking facilities with profile details, for instance to pick the first opaque voxel within a volume. Application API for data access and sharing, allowing data management and processing Built-in SEG-Y reader handles 2D or 3D data, 8/16/32 bit integer or IEEE/IBM floating point. Quality/accuracy control: slicing, interpolation, on-gpu gradient-based lighting, floating point buffers, deep textures, bump mapping on slices 5 The main concepts such as classification, compositing, interpolating, lighting, slicing, region of interest and clipping are supported as a set of classes that are fully integrated with the Open Inventor scene graph. The application programming interface is intentionally compact, yet highly flexible and extensible, in order to make the programmer's task as easy as possible. As proof of its simplicity, rendering a 3D data set minimally requires instantiation of only two classes: one for describing the data and one for drawing the volume 6. VolumeViz has been designed to take full advantage of 3D graphics or specialized volume rendering hardware, while at the same time providing the highest possible image quality. VolumeViz automatically manages system memory and texture memory, allowing large volumes to be rendered efficiently. Just as simply, the Large Data Management feature allows extremely large volumes to be navigated freely on standard hardware. ROAMING HUNDREDS OF GIGABYTES EVEN ON A LAPTOP WITH VOLUMEVIZ LARGE DATA MANAGEMENT Today s seismic data sets are commonly 10's or even 100's of gigabytes. This easily exceeds the available system memory on 4 VolumeViz 6.0 5 VolumeViz 6.0 6 A greyscale transfer function is used by default standard hardware, and far exceeds texture memory even on high-end graphics hardware (still less than 1 GB). Even moderate sized 3D or extended 2D datasets can exceed standard resources, especially if you consider the multiple datasets and attributes that are involved in typical workflows. The VolumeViz solution is a hierarchical, multi-resolution Large Data Manager (LDM). This innovative technology allows rendering of extremely large data sets with reliable interactive navigation even on relatively low-end machines such as laptop with 3D graphics. VolumeViz LDM is included in commercial applications today and has been stress-tested on hundreds of gigabytes of data by key customers. It is designed to support up to multiple terabytes of data! The VolumeViz LDM solution combines new, highly efficient algorithms for managing system and texture memory with highly optimized rendering code developed over five previous generations of VolumeViz. The default data manager uses a spatial representation adapted to extremely large and possibly very asymmetric volumes. The LDM software architecture also allows this and other components to be replaced for customization and further enhancements. EFFICIENTLY MANAGING MULTI-RESOLUTION DATA For best results the data is pre-processed to create a hierarchy of lower-resolution, sub-sampled "bricks". The bricks at each higher level represent a larger portion of the total volume at a lower resolution. So overall the top level brick contains a low resolution "overview" of the entire volume and the lowest level bricks contain the full resolution data. This pre-processing requires 15-20% additional storage for the volume data. An optimized converter 7 is provided to efficiently generate multi-resolution data from input files using the built-in SEG-Y reader or any other format through an application-defined reader. It can be used either standalone or integrated within your application. The converter is being parallelized for even faster processing on SMP 8 systems or clusters 9. The LDM multi-resolution format is optimized for fast access to large volumes, but you may also override LDM with your own custom or legacy multi-resolution format or on-the-fly sampling. UNRESTRICTED ROAMING IN VOLUMES Generally the top-level brick is loaded first, providing almost immediate visualization of the volume (or any portion of the volume). Then, the image is automatically refined by loading higher resolution data and higher resolution textures. Interactivity is maintained because rendering always uses data already loaded asynchronously into memory. Although lower resolution data may be displayed temporarily, maximum image 7 Performance is multiplied by 3 in release 5.1 8 Symmetric Multi-Processing 9 VolumeViz 6.0 2 Open Inventor and VolumeViz LDM Cluster Edition

quality is reached quickly. The allocation of system resources depends on the current view, current objects (slices, sub-volumes, etc.) and a set of automatic and user-controlled heuristics such as "slices have higher priority than sub-volumes". In order to maintain interactivity, VolumeViz LDM takes into account a variety of constraints including system memory, texture memory, rendering speed, data loading speed (disk to system memory) and texture loading speed. Even better, the Large Data Manager can automatically pre-fetch the necessary data 10. By loading a hull around the geometry or region of interest, LDM can anticipate user interaction: when a slice or ROI moves, the data has already been loaded in memory, and users immediately see the higher resolution image. The default heuristics for culling, geometry, volume, and ROI are designed to give the best user experience in most cases and can be easily overridden by the application for best user interaction. low-level control of graphics rendering. The high flexibility of the recent generation of graphics processors enables real-time visualization effects that can dramatically improve data visualization, and also offer promising opportunities for cost-effective, high-performance data processing. VolumeViz automatically takes advantage of the latest graphics hardware features such as programmable shaders and advanced texture support 11 : Compressed textures for extending texture memory capacity and accelerating transfers. Bump mapping effect on slices to emphasize gradient differences in data. Render to floating point RGBA buffer, to reduce image accumulation errors and achieve highest quality. Load data or colormap in deeper textures, to prevent loss of precision for instance with 16-bit data or 12-bit colormaps. Data min/max and advanced transfer functions, for most flexible real-time management of color and opacity mapping. GPU-based isosurface or attribute rendering, for fast interactive segmentation of large data. SCALABILITY AND LOAD BALANCING Data is loaded asynchronously using several separate threads, taking advantage of multi-processor systems or RAID storage systems. Performance and image quality automatically increase as system constraints are raised, for example more system memory, next generation graphics board, a faster disk drive or an additional CPU or cluster node. In summary, VolumeViz LDM allows unprecedented access to extremely large data sets and at the same time provides a remarkably interactive user experience. The user perceives that any portion of the volume can be displayed immediately, the volume can be navigated smoothly at all times, slices and probes are responsive at all times and image quality is automatically managed to the best possible level using available resources, with no performance drop. EXPLOITING NEXT GENERATION GRAPHICS HARDWARE Graphics hardware has evolved into high-performance Graphics Processing Units (GPUs). Shading languages as illustrated by Nvidia s CG have been made available to the programmer for 10 VolumeViz 6.0 Furthermore, Open Inventor 5 introduced easy integration of vertex and fragment shader programs in applications, with powerful and consistent insertion within the scene graph, shader-language independence, straightforward use in existing applications, consistent persistence and exchange with Open Inventor file format, built-in shader programs and examples. Open Inventor provides the best framework to effectively manage advanced GPU rendering and computing. FROM VISUALIZATION TO DATA MANAGEMENT VolumeViz provides a complete data access API allowing applications to take advantage of Large Data Management for accessing data associated with a sub-volume, plane, point or polyline 12 at arbitrary resolution. This makes VolumeViz not only a visualization tool but a powerful middleware for volume data management. MANAGING MULTI-DATA AND ON-THE-FLY ATTRIBUTES 13 Attributes derived from the data can be pre-computed or computed on-the-fly using the CPU or GPU. With multi-data support, VolumeViz provides comprehensive support for flexible strategies: Data transforming This is a one-to-many scenario, where a volume managed by LDM can be rendered as one or multiple derived volumes, for 11 VolumeViz 6.0 12 VolumeViz 5.0.4 13 VolumeViz 6.0 3 Open Inventor and VolumeViz LDM Cluster Edition

example, computing a seismic attribute volume on-the-fly. Derived tiles are managed in memory, but this derived data does not exist on disk and derived data only needs to be computed for tiles that are actually needed. Data combining Here multiple volumes managed by LDM are rendered as one volume ( many-to-one scenario): source volumes are managed in synchronization. An example is the computation of volume difference on-the-fly. Again the derived tiles are managed in memory, do not exist on disk and are only computed when needed. Render combining In this scenario, multiple volumes and multiple colormaps (or other data) are combined by the GPU using fragment shader programs. The CPU is freed from such computation. This could be used for instance for clipping against a mask volume, for multi-channel rendering and for advanced multi-dimensional transfer functions for better highlighting dataset features. VISUALIZATION SERVER FOR LARGE DATA WITH REMOTE RENDERING Oil & gas exploration and production is a global business involving massive and sensitive data. How to avoid costly or unsafe data transfers? How to take advantage of remote high-performance hardware resources such as servers, graphics, memory, storage? How to increase return on investment by better sharing such systems? How to facilitate collaboration of geographically dispersed team members? Remote rendering is a built-in feature of Open Inventor from Mercury, which allows an application to run on a remote "server" machine and also do 3D rendering on the remote machine, while displaying on a local machine. The power of Remote Rendering is that it allows users to view and interact with very large data sets from almost any client machine located almost anywhere. Remote Rendering allows a single graphics server to provide visualization for multiple client machines. Local client machines do not need any 3D graphics capability because rendering is handled completely on the remote machine. The local client machine only needs the ability to display images. In some cases multiple client machines can connect to the same remote session, enabling collaborative visualization without the need to share application data between the client machines. Open Inventor remote rendering runs over standard fast networks. The client machine may be running any remote viewing tool, preferably combined with a server transmitting compressed images to reduce network bandwidth requirements. Remote rendering does not require any change to existing Open Inventor applications. It can be enabled simply by setting an environment variable. When the capabilities of evolving graphics hardware are combined with the powerful system architectures on which Mercury is working, customers can expect to see in the near future major improvements in solutions for sharing and remotely accessing large data sets. EYES WIDE OPEN ON COMPLEX DATA Geophysicists have experienced an increase of efficiency brought by improvements in display size and resolution. However when geophysicists have to analyze very large volumes, even high-resolution dual-screen workstations make too limited use of human perception. Texaco was among the first companies to show how large immersive displays could dramatically improve collaborative work. Today, visualization theaters, typically using three projectors and high-end graphics hardware, are commonly used for collaboration, interpretation, modeling and planning. Total recently focused on an additional concern. Interpreting seismic data requires correlating information with tiny details across very large scales within acquisitions ideally 1 sample value should correspond to 1 pixel, what some call scale 1 seismic visualization. Therefore even higher resolution (not only larger size) is needed for displays, requiring a tiled array of projectors or screens. The only way to display this data with such high resolution is to use a large logical display composed of many individual displays arranged in an array. Moreover one may wish stereoscopic display, which greatly helps depth perception in particular with volume rendering and is mandatory for immersive VR applications. This puts a heavy load on graphics hardware, especially if you consider the trend for increasing data size and complexity which some estimates say is doubling every year in the oil & gas industry. MULTI-PIPE SYSTEMS Graphics scalability was first addressed by incorporating multiple graphics engines ( pipes ) in multi-processor systems with fast interconnects enabling shared memory. While such specialized high-end SMP systems have proved to be mandatory for a number of demanding problems, they now face tough competition from the quick evolution of technology, performance and cost of mainstream commodity hardware. Since release 3.0, the Open Inventor Multipipe extension automatically manages multi-threaded parallel rendering for multiple displays. It is designed to optimally support such architectures with little effort for the developer. This will prove even more useful with the new generation of multipipe systems, now comprising PCs with two graphic boards 4 Open Inventor and VolumeViz LDM Cluster Edition

on high-bandwidth PCI Express connections. Handling multipipe at the scene graph level provides the best possible performance. However, the number of video outputs as well as the overall system power/bandwidth of dual-graphics PCs is still too limited to address the demanding requirements of oil & gas users: combining the power of multiple PCs is still required. PC CLUSTERS FOR GRAPHICS Computing clusters have been used for a long time and are widely used in the oil & gas industry for data processing and simulation. They should not be confused with graphics or visualization clusters, made of multiple inter-networked graphics nodes. Graphics clusters have been identified as a cost-effective alternative to high-end multi-pipe systems in order to efficiently drive multiple displays in parallel and maintain or increase performance compared to a standard single display. Beyond research and academic VR usage, can they truly fit the demanding requirement of oil exploration? In fact, cluster solutions for oil & gas data visualization are now reaching maturity, provided that the right software architecture is used and carefully tuned with well chosen hardware. Disappointment may result from choosing too generic a solution that is not tailored to the application s needs, not robust enough or that only addresses rendering requirements without taking into account large data requirements. Real expertise is required for designing a solution taking full advantage of the hardware resources. In cooperation with Total, which has initiated a project addressing PC cluster architectures and large displays, Mercury has designed the most effective tools available to developers for building working solutions. WHICH SOFTWARE DESIGN? As a possible taxonomy, three main strategies for cluster rendering can be distinguished, with different trade-offs. For the oil & gas industry it is critical to consider the implications for managing very large volume data, not just the rendering performance. Generally a large volume will not fit in system memory, so we have to consider the available system memory as a cache. Distributed OpenGL The application runs on a single cluster node (master). OpenGL commands are intercepted and sent to slave nodes where actual rendering is done, depending on some defined display configuration. Dedicated middleware such as the Chromium open source research project has been designed to distribute OpenGL across the network more efficiently than OpenGL GLX protocol, and also manage display configurations and advanced flow control. Such transparent solutions do not require much, if any, change to the application and can potentially support any existing OpenGL application without change (including Open Inventor-based applications). In practice, robustness and interoperability issues limit adoption. Even optimized, compressed, multicast and cached, the OpenGL throughput will still be limited by network bandwidth. Smart use of OpenGL - as done by Open Inventor - can significantly reduce the bandwidth problem for some applications such as walkthrough or design reviews with static models or rigid animations, but there is no easy solution for heavily changing data or interactive visualization of massive data. The whole data flow to be visualized must be input, processed and transmitted by the master node. So this solution does not make much use of the CPU and system memory on each slave node. Worse, the total system memory available for caching volume data is limited to what can be installed on the master node. Distributed application A copy of the application runs on each node of the cluster (master and slaves). The master synchronizes the slave data state by replicating user input to slaves and also synchronizes the slave rendering. The clear benefit is to minimize network traffic, which makes this approach effective in many cases, at least with moderately sized data. This approach has been used for example in Mercury s AmiraVR application and in Open Inventor applications combined with CAVELib from VRCO. However, such an application-specific solution is much more difficult for developers, who have to specifically design the application for distribution. This can be especially difficult for existing applications. Each application instance must load all the data, so the network bandwidth saved is now lost loading data onto each node, despite file sharing and caching. This approach also wastes computing power by almost exactly duplicating application execution on each slave. So we could say that this approach makes too much use of the CPU and system memory on each slave node (for no net gain) and the total system memory available for caching volume data is still limited to what is installed on any single node. Distributed scene graph The application runs on a single cluster node (master) and a small "agent" program runs on each slave node, maintaining and rendering a copy of the scene graph. During execution the master monitors changes to the scene graph and only these high level changes are transmitted to each slave. This approach combines the best features of the previous two. Similar to distributed OpenGL, distribution of the scene graph can be transparent to the application. But similar to the distributed application strategy, very little network bandwidth is required. Plus each slave node is traversing a high level scene graph, rather than replaying low level OpenGL commands, so more sophisticated optimizations are possible. Like the distributed OpenGL approach, data read directly by the application is only needed on the master node and application specific computing is not duplicated on the slave nodes. This makes the CPU on each slave node available to optimize traversal of the scene graph specifically for its portion 5 Open Inventor and VolumeViz LDM Cluster Edition

of the view volume. And this, in turn, leads to better performance and load balancing, almost without development effort. There is some inefficiency in the use of system memory on the slave nodes if each node has a copy of the contents of the scene graph. However the real advantage of this approach becomes clear when we consider very large volume data. Unlike geometry, volume data is not being stored in the scene graph and so will not be duplicated on each slave node. It should be possible for each slave node to load only the voxels needed for its portion of the view volume, so the total system memory available for caching volume data can be approximately the sum of all the memory on all the slave nodes. OPEN INVENTOR CLUSTER EDITION A TRANSPAR- ENT DISTRIBUTED SCENE GRAPH Open Inventor Cluster Edition is based on a distributed scene graph strategy: the application simply runs unchanged on the master node, and its scene graph is optimally distributed to the slave nodes, where render agents manage the parallelized rendering of portions of the total image. The master node manages scene graph synchronization with compact serialization and also manages synchronization of rendering. A communication layer is included supporting different protocols from TCP/IP to fast Infiniband. To deploy the application, once appropriate standard packages are installed on the cluster, you just install the application on the master node. Further upgrades of applications will only require update of the master. The display configuration can be defined by a simple file or programmatically, similar to the Open Inventor Multipipe extension. Automatic distribution of application objects with custom scene graph nodes Open Inventor is very flexible and encourages extensibility. While most of the time developers will just pick among the large set of existing classes, they can also conveniently extend the scene graph with application objects. These custom scene graph nodes can even be packaged as shared libraries and then dynamically loaded automatically. Extending this mechanism, the master can automatically distribute the application-specific extensions to the slave Open Inventor Render Unit. For maximum flexibility, an API is also provided to exchange application messages and data between master and slaves. Combining cluster and multipipe Cluster and multipipe approaches are not exclusive: cluster nodes can be for instance a PC with two graphics boards. Very compact clusters can therefore give higher performance. Open Inventor 14 can support combined use of multi-pipe nodes within a cluster. Aggregating power with image compositing Graphics clusters can actually do more than just maintain performance when increasing the number of displays and the resolution (display scalability). The computation and graphics power of several cluster nodes can be combined to manage one individual display, accelerating the performance. Open Inventor Cluster Edition 15 can support hardware or software composition, for a wide range of high-performance solutions tailored to the application: sort-first or sort-last composition, multi-display, single display, remote rendering and visual servers. OPTIMAL SCALABILITY WITH VOLUMEVIZ LDM VolumeViz LDM has been specifically optimized for cluster rendering using a distributed scene graph strategy, with all the advantages mentioned previously. Which turns Open Inventor into an optimal middleware for the management and visualization of very large volume data. Each slave node loads only the tiles it needs (for instance, based on view frustum culling). So all the slave nodes are working truly in parallel, each (approximately) loading a different set of tiles. The largest volume that can be loaded (cached) in system memory is now the sum of the system memory on all the slave nodes. Each slave is independently managing its graphics board(s), so the total number of voxels that can be loaded in texture memory is now the sum of the texture memory on all the slave nodes. Finally, if derived volumes, for example seismic attributes (see the Managing multi-data section), are being computed onthe-fly, that computation will be done in parallel, distributed across the cluster. Therefore all the resources of the slave nodes, memory and CPU, can be fully exploited. 14 VolumeViz 6.0 15 VolumeViz 5.0.4 6 Open Inventor and VolumeViz LDM Cluster Edition

SCALABLE INTERACTION FROM DESKTOP TO VR SYSTEM Seismic interpretation, reservoir modeling and simulation and other geoscience tasks typically involve very complex scenes combining large volumes and slices with large geometries for faults and horizons, complex meshes, etc. Applications need to offer users an efficient way to interact with these data. Fast edit mode for interaction with large scenes Users want to smoothly manipulate probes, 3D cursors or other interactors without being constrained by the rendering performance of large scenes. For this purpose, Open Inventor supports fast rendering update of local changes in complex scenes, without requiring re-rendering the full scene. This feature is very easy to use in applications and is highly valuable when you need to frequently modify a relatively small part of a large scene that cannot be rendered at interactive rates. Easily make your application VR aware Immersive VR systems with 3D interaction are the ultimate facilities for highly interactive applications. The Open Inventor suite helps you to create, migrate, and cost-effectively maintain a VRaware application. The Open Inventor VR package takes care of display configurations, parallelized rendering, head-tracked visualization, 3D device events and interaction with the scene. Through simple configuration files, Open Inventor can support multi-monitor desktops, panoramic displays, immersive workbenches, high-resolution image walls, VR rooms... For controlling applications from the immersive system, the Open Inventor DialogViz package allows a single GUI definition to be used for the user interface for both desktop and VR systems. The Open Inventor DialogViz package extends the scene graph with a comprehensive set of GUI components as nodes. The user interface is simply described in a resource file using the Open Inventor format (VRML-like). The desktop version of the user interface uses standard 2D widgets and dialog boxes, while the immersive user interface uses 3D components with a look and feel that can be easily customized and comes with several skins. While Open Inventor can be used completely standalone for VR applications, it is also interoperable with your preferred or legacy VR support tools. A breakthrough for high performance visualization Compared to other approaches, Open Inventor Cluster Edition and VolumeViz LDM provide, in a transparent way, the best possible load balancing and scalability with any available combination of CPUs, GPUs, channels, memory, storage, and network resources, thanks to the high level of data synchronization, caching and management. Open Inventor Cluster Edition is designed for transparent adaptation of any existing Open Inventor and/or VolumeViz LDM applications. No effort is required for user interface integration and deployment on arbitrary configurations. Moreover, Open Inventor s ease of use, flexibility and interoperability makes it the perfect tool for easily creating new cluster-aware applications or for extending legacy applications. A complete framework to improve your productivity Real-life oil & gas applications must provide a wide range of visual representations and interaction tools, involving tedious details and advanced techniques which can mean considerable time and risk to implement from scratch. With an easy learning curve for your development team, the high-level Open Inventor suite offers more than 1000 ready-foruse classes integrated in a user-friendly framework for rapid development that lets you focus on your real business: Ready-for-use viewers, interactors, behaviors MeshViz classes for data visualization: comprehensive scientific data representations on 2D/3D meshes, visualization for reservoir models, flows, CAE simulation results, business graphics and statistics, contouring, faults Built-in performance optimizers, Large model visualization, advanced culling VR and collaboration, multi-thread, multi-pipe, remote rendering... Shader integration for advanced rendering (ARB, CG, GLSL) High-resolution image output, fast high-quality 3D vector rendering to PostScript, HPGL, GDI/EMF, CGM Support for VRML, NURBS, CATIA, STEP, IGES, OpenFlight The scene graph paradigm provides ready-for-use graphics programming patterns. The object-oriented design encourages extensibility and customization to satisfy specific requirements. ALL-TERRAIN TOOLS FITTING YOUR ENVIRONMENT Open Inventor from Mercury is the de-facto industry standard, bringing true hardware independence and covering a full range of systems including: UNIX, Linux, Windows, and Mac OS X, for 32 or 64 bits. Mercury has also tuned Open Inventor for interoperability with your preferred or legacy environment. For instance you can even take advantage of Open Inventor with existing OpenGL code, or extend it at the OpenGL level. Open Inventor provides GUI components for Win32, MFC, X11/Motif, Qt, and Java that can be customized to fit your specific constraints. 7 Open Inventor and VolumeViz LDM Cluster Edition

Open Inventor also supports development in multiple languages, including C++ and Java. Open Inventor 6.0 will also be integrated with Microsoft s.net framework to provide a multi-language development solution. FUTURE-PROOF DEVELOPMENT Supporting key customers in oil & gas for many years, Open Inventor from Mercury is proven to be the safe choice for the long term and the most flexible tool to transfer technology evolutions and unique innovations to your solutions. For instance, Open Inventor-based applications benefit from improvements coming from evolution of the OpenGL API in the simplest way, often without any code change. Extensions of the API and new class modules are carefully designed to introduce powerful new capabilities for your application in the most simple, transparent and consistent way, protecting your investment and anticipating needs that you may not even foresee. Last, the interoperability and extensibility ensure your complete freedom to best adapt the toolkit to your specific needs. We don t just solve issues for today, but provide long-term solutions based on the wisdom of experience. JOIN A PARTNER COMMITTED TO YOUR PROJECT SUCCESS Dedicated to servicing OEM partners, Mercury brings more than 20 years experience in 3D visualization. Our support team pays particular attention to constraints of professional developers, working closely with R&D for best phasing with your development schedule through road-mapped patch releases, beta versions and major releases, as well as quick assistance for critical issues. Furthermore, the Mercury Professional Services team is at your disposal to increase your efficiency through training, expertise or custom development covering the whole life cycle of your project: software and hardware requirements, specifications, prototyping, migration assistance, code expertise and optimization, custom component development, specific product extensions, support for system deployment and even cooperative research and development. MORE BRICKS FOR BUILDING SOLUTIONS Beyond Open Inventor, Mercury provides a full range of software or hardware components that can help you to create added value solutions for the oil & gas industry: SAL Scientific Algorithm Library, for signal and image processing OpenRT scaleable real-time ray-tracing engine Amira high-level framework for 3D material visualization and analysis XBi series high-density computing/graphics nodes for compact clusters, for instance for onboard 3D visualization Dual Cell-Based Blade significantly improves performance for graphic-intensive workloads and computationally intensive high-performance computing (HPC) applications in seismic processing Open Inventor is a registered trademark of Silicon Graphics, Inc. Mercury Computer Systems, Inc. is a source licensee of OpenGL and Open Inventor from Silicon Graphics, Inc. VolumeViz and MeshViz are trademark of Mercury Computer Systems, Inc. Amira is a registered trademark of Konrad-Zuse-Zentrum fur Informationstechnik Berlin (ZIB). Other products mentioned may be trademarks or registered trademarks of their respective holders. Mercury believes this information is accurate as of its publication date and is not responsible for any inadvertent errors. The information contained herein is subject to change without notice. Ownership of Materials This document, including layout and graphics, is a copyrighted work and is protected by worldwide copyright laws and treaties. You may not reproduce this work, prepare derivative works based on this work, distribute copies of this work to the public by sale or other transfer of ownership, or by rental, lease or lending, perform or display this work publicly, or otherwise commercially exploit this work. Copyright 2005 Mercury Computer Systems, Inc. For more information, go to www.mc.com 372-0905-openinvvolvizwp 8 Open Inventor and VolumeViz LDM Cluster Edition North America 199 Riverneck Road Chelmsford, MA 01824-2820 USA 978-967-1401 866-627-6951 Fax 978-256-3599 Asia No. 2 Gotanda Fujikoshi Bldg. 4F 5-23-1 Higashi Gotanda Shinagawa-ku, Tokyo 141-0022 JAPAN +81 (0) 3 5420 3881 Europe Immeuble Le Montreal 19 bis, avenue du Quebec Villebon-sur-Yvette 91951 Courtaboeuf Cedex FRANCE +33 (0) 1 69 59 94 00