1. Scalable Science on the Web? Challenges and Possibilities Don Brutzman Modeling, Virtual Environments and Simulation (MOVES) Naval Postgraduate School, Monterey California brutzman@nps.navy.mil NSF Workshop: Grand Challenges eScience

2. 5 December 2001THE MOVES INSTITUTE Two topics (rants) Scientific method, modeling & simulation Proposed “grand challenge” for Science on Web Enabling technologies 3D, XML languages, behaviors, networking, physics aka large-scale virtual environments

3. 5 December 2001THE MOVES INSTITUTE Some definitions Model Representation of process in real world Simulation Behavior of a model over time

4. 5 December 2001THE MOVES INSTITUTE Scientific method For many hundreds of years, scientific method has essentially been repetition of two steps:  Theory  Experiment However, two virtual analogs now exist:  Modeling  Simulation

5. 5 December 2001THE MOVES INSTITUTE Typical process: hypothesize, test, repeat  Theory  Experiment  Modeling  Simulation Process of scientific inquiry

6. 5 December 2001THE MOVES INSTITUTE Actual practice more often a combination of each:  Theory  Experiment  Modeling  Simulation Process of scientific inquiry

7. 5 December 2001THE MOVES INSTITUTE Simulation advantages over experimentation Repeatable, adjustable, low cost or “free” Can insert various error distributions Zero-error perfect case for algorithm correctness Statistically definable for measuring variations, rigor Can be intentionally extreme to test robustness Can predict otherwise-impossible conditions Catastrophic failure (of simulated system) is OK

8. 5 December 2001THE MOVES INSTITUTE Simulation complementing experimentation Forward: can sometimes insert experimental data into simulations Mix of both needed for Verification (computationally stable) Validation (predictions match measured results) Reverse: can sometimes insert simulation data into experiments

9. 5 December 2001THE MOVES INSTITUTE Most Ignored Word in Computer Science “Science” How many computer scientists run experiments? Fairly widespread problem / occupational hazard Try looking for Experimental Results section in conference, journal papers Most other disciplines won’t publish without results Hmm, what about Simulation Results sections? and the answer is…

10. 5 December 2001THE MOVES INSTITUTE Science characteristics Theories and models tend to be disjoint or at least disconnected Assumptions, limitations and inputs of one model tend to be outputs of another model Conjectural, but experts tend to know how contributions in their field all fit together Biggest challenges are often cross-disciplinary

11. 5 December 2001THE MOVES INSTITUTE Science characteristics Good experimental data is usually available for theoretical models At least within limited ranges of experiments Not usually available, though (despite NSF efforts) Simulation results crucial to conducting science but simulations are rarely reported, published, linked or re-used Interchangeability of simulations and experiments is not supported

12. 5 December 2001THE MOVES INSTITUTE proposed Grand Challenge in e-Science Enable scalable interconnection of Science on the Web, using theoretical models, experimental results and simulation results.

13. 5 December 2001THE MOVES INSTITUTE Enabling technologies XML schemas for Scientific languages: MathML, Chemistry ML, etc. Others possible, even experiment-specific schema Integration feasible through XML namespaces Metadata Dublin Core, Resource Description Framework (RDF) Semantic Web enables agents and other processes Internationalization (i18n) and Localization We also live on planet Earth, not just in U.S.A.

14. 5 December 2001THE MOVES INSTITUTE Enabling technologies X3D graphics: Web interchange for 3D models model composition occurs in virtual environments Web-adept integration with other XML languages Behavior protocols So scenes, models, humans etc. (i.e. applications) can interact Networking infrastructure Client, server, peer-to-peer, monitoring, services

15. 5 December 2001THE MOVES INSTITUTE Extensible 3D (X3D) X3D graphics: Web interchange for 3D models Virtual Reality Modeling Language (VRML) 3 rd generation ISO standard with XML encoding 3D render hardware already deployed everywhere Get 3D models “out of box,” out of proprietary islands http://www.web3D.org Deliverables:  Specification  Tagset  API  Authoring tools  Content  Conformance

16. 5 December 2001THE MOVES INSTITUTE Configuring Powerpoint for 3D Takes a few minutes of configuration to set up: http://web.nps.navy.mil/~brutzman/Savage/ InstallingCortonaBrowserAsPowerpointControl.ppt http://web.nps.navy.mil/~brutzman/Savage/ InstallingCortonaBrowserAsPowerpointControl.ppt http://web.nps.navy.mil/~brutzman/Savage/ InstallingCortonaBrowserAsPowerpointControl.html http://web.nps.navy.mil/~brutzman/Savage/ InstallingCortonaBrowserAsPowerpointControl.html

17. 5 December 2001THE MOVES INSTITUTE online X3D/VRML example: gimbals [go to full-screen Presentation mode to activate] [PgUp/PgDn to change viewpoints, arrow keys or mouse to rotate]

18. 5 December 2001THE MOVES INSTITUTE online X3D/VRML example: kelp forest [go to full-screen Presentation mode to activate] [PgUp/PgDn to change viewpoints, arrow keys or mouse to rotate]

19. 5 December 2001THE MOVES INSTITUTE 3D myths, enablers File size is big Actually much smaller than video/audio, with added benefits of interactivity and viewpoint independence Modeling is hard Data-driven autogeneration using templates works “Content is king” Navigation interfaces are klunky Yes, sorta like hypermedia prior to NCSA Mosaic

20. 5 December 2001THE MOVES INSTITUTE A simple challenge? Goal: Clearly demonstrate XML language interoperability Example: Collaborative visualization for cardiac diagnosis  XHTML: hypermedia web pages  SVG: Scalable Vector Graphics 2D diagrams  SMIL: Synchronized Multimedia Interface Language streams  MathML: biomechanical, biochemical models  X3D: visualize changes to 3D model of heart

21. 5 December 2001THE MOVES INSTITUTE Behavior protocols Highly specialized application-level protocols Perhaps unique to each type of model Examples: IEEE Distributed Interactive Simulation (DIS) protocol W3C XML Protocol (XP) work, SOAP, others NPS Dynamic Behavior Protocol  XML-defined packet payload, can modify/replace at run time

22. 5 December 2001THE MOVES INSTITUTE Network considerations, needs Client operations: applications, obviously Server operations: needed but typically blocked Multicast: multiple interactions simultaneously Scalable peer-to-peer communications Area of interest management (AOIM) Robust fallback to unicast tunneling Network monitoring Controlled, repeatable experimental environment Repeatability more important than strict causality Much bigger than 2-point optimization

23. 5 December 2001THE MOVES INSTITUTE Network considerations, needs Support services NTP for clock synchronization LDAP for directory/discovery services, e.g. VRDNS Security for signing, authenticity, etc. Repositories and archives of interoperable content Common theme: “middleware solutions” needed but framework is the enabler, not a legislative end goal Forcing function/goal: growth, composability and scalability matching the capabilities and growth patterns of Web Push all the way to desktops, not just infrastructure

24. 5 December 2001THE MOVES INSTITUTE Some physics considerations Physics of interactions between models needed Important part of VR is reality, not virtual Some intractable problems are yielding e.g. N-N collision detection appears tractable using variable-resolution algorithms + network partitioning Shared supercomputer problems, solutions Typically low-resolution physics on clients, then high-res physics on servers as shared resource Good application area for reliable multicast

25. 5 December 2001THE MOVES INSTITUTE Some physics considerations Contrast in disciplines Operations Research (OR) has rigorously consistent mathematical notation, definitions Mechanical Engineering (ME) hydrodynamics doesn’t  Progress is much harder to validate, repeat Probably typical situations for other sciences too Backdrop of “real world” data has caught up Terrain, satellite imagery, remote sensing, etc. etc. Needs to be available on demand as initial conditions, bounding assumptions, model/simulation/experimental data in its own right

26. 5 December 2001THE MOVES INSTITUTE proposed Grand Challenge in e-Science (reprised) Enable scalable interconnection of Science on the Web, using theoretical models, experimental results and simulation results. Web 3D virtual environments are where these capabilities will be most needed and most visible.

27. 5 December 2001THE MOVES INSTITUTE Contact Don Brutzman brutzman@nps.navy.mil http://web.nps.navy.mil/~brutzman Code UW/Br, Naval Postgraduate School Monterey California 93943-5000 USA 831.656.2149 voice 831.656.3679 fax