Interactive Simulation of Contaminant Evolution through Porous Media.
Project Team: Jon Goldman , Louis Rossi, George Sohos and Rob Stevenson.
Last modified: February, 6 1996
These are oil saturation isosurfaces for a contaminant leak into a porous cube.
Purpose:
We hope to demonstrate the capabilities of leading edge parallel computers and numerical methods coupled to the fastest network links can benefit the scientific community and ultimately science. In particular, we will simulate motion of benzene, an environmental contaminant, through a cube of soil and water. This simulation requires considerable computational resources available at a limited number of sites around the United States. However, with high speed computer networks, the data can be relayed to remote sites around the country. In this case, the simulation ran at the National Center for Supercomputer Applications but was visualized with CAVE virtual reality equipment in San Diego at the Supercomputing '95 conference.
Broad Description:
When chemical contaminants are accidentally or intentionally poured onto the ground, they flow downward and can diffuse into groundwater supplies. For obvious reasons, it is not practical to experimentally measure the flow of these contaminants through the soil. Thus, there is a desperate need for models and simulations of contaminant flow so that scientists and policy-makers can accurately assess risks associated with faulty contaminant storage and plan effective cleanup strategies when spills do occur. Our conference demonstration is an attempt to build a tool that will will guide environmental policy and cleanup projects by allowing the user to measure contaminant flow through a porous cube.
Our simulation is a finite difference code simulated on a parallel machine. We simulated TCE (trichloro-ethylene), water and air. The computer simulates the evolution of saturations and pressures through the cube with variable porosity. For demonstration purposes we hoped to attain a resolution of 40^3, but network limitations forced us to restrict ourselves to 20^3.
Inside the CAVE, the users passively observe the flow of water and TCE through the cube, and interact with the physical model to alter the permeability of the cube and add sources and sinks as well as water sources and sinks. Also, the user could adjust the visualization parameters such as isosurfaces of saturations in the flow.
The particulars...
There are three major components to our environmental simulation: the numerical engine, the high-speed network connection and the CAVE user interface.
The numerical engine.
At the center of our project is the numerical engine. We have taken a mathematical model for the motion of multiphase flow (more than one fluid or gas the same area) and discretized it onto a grid. The model takes the form of several partial differential equations (PDEs). One of them expresses conservation of mass. The other relates the mass flux through a given region to a velocity and diffusive term. The numerical engine approximates the solution to these PDEs at a given moment, and then marches forward in time. Doing this fast enough for real-time interaction requires a supercomputer. In this case, we used the NCSA Power Challenge Array to drive our simulation.
The high speed network.
Since the numerical engine is in Illinois and the conference site is in San Diego, the information about the state of the porous cube must be transmitted along computer network lines. Ordinarily, computer network lines are two slow to support real-time animation at a rate of three to fifteen frames per second. However, the Global Information Infrastructure (GII) Testbed has made available to us special high-speed communication lines for use at this conference. Using a message passing protocol known as Message Passing Interface (MPI) , our programs communicated with one another along these fast communication lines.
The CAVE virtual environment.
The CAVE is a virtual environment consisting of three walls and a floor, a pair of goggles and a wand. The goggles allow the user to view stereoscopic images projected onto the floor and walls giving the user a three-dimensional view of our porous cube. The wand is a three dimensional analog to a mouse. Within the CAVE, the user can view the cube of any point of view, adjust the oil or water isosurfaces and change simulation parameters. If effect, the users had full control over the numerical engine from two thousand miles away. In the future, scientists in the CAVE could measure the affects of contamination and even attempt different cleanup strategies.
Results:
We successfully demonstrated our interactive porous media simulation at Supercomputing '95. Visitors could view the evolution of contaminants through a variety of porous cubes. Some of the data was precomputed and some was computed in real time for novel configurations while visitors watched. Since there were a large number of demos, our team used the ImmersaDesk rather than the CAVE. Though this hindered the visualization experience slightly, the visitors did a good feel for what happens as contaminants leak into the ground. Click here for an example of our work.
For most of the demo period, the special high speed IWAY network was not working, forcing us to improvise. Fortunately, we thought this might happen and MPI is flexible enough to allow us to change platforms and networks without major modifications. When the IWAY network was not functioning, we simply ran our simulation locally on the less powerful machines at the conference site and passed messages over the Internet. When the IWAY came back up, we ran our simulation on the faster NCSA machines over the vBNS and ATM links. The speedup was remarkable.
Lessons learned:
The project team has answered some of the following questions in the hopes that it will help others pursuing similar endeavors.
Was the GII Testbed successful?
Short answer: Yes and no. Some things went well. Others did not. Read on to learn about the specifics.
Yes, it was a success. At the end of the day, many teams, including this one, took advantage of vBNS and ATM networks to link high speed computing with state-of-the-art virtual reality visualization systems. The user interface was friendly. Naive visitors could get a feel for the science. Experts could use our software as a tool for environmental modeling.
No, it was a failure. First, there is no denying the fact that the IWAY network was down most of the time. When it was up, it was not reliable in the sense that when it often went down without warning. Second, there were too many projects that had almost nothing to do with networking or high performance computing. The purpose of the GII Testbed was to demonstrate the prospect of connecting supercomputers and virtual reality visualization with high speed networks. Projects that failed to do this, diluted the demonstration session and obscured the main point. Either there was not enough refereeing at from the onset, or there were not enough checkpoints along the way to determine which projects were going to meet the Testbed criteria.
Were we successful?
Heck yes. We delivered a project that combined supercomputing, visualization and high speed networking.
Will anyone ever use our software again?
Since CAVEs, Immersadesks and the like are expensive, it would require somehow making VR more affordable for modest research facilities. Some alternatives include big computer screens with stereo goggles. We were approached by some high school teachers at the conference who are considering some sort of VR for their classrooms.
What is the future of VR and GII networking challenges?
This is hard to say. It will depend in a large part on funding priorities in the government and with industry. In the conference program, I (LFR) did not see any projects that were producing publishable results. I think some funding should go to a long term project with tangible results of scientific merit. If successful, it would probably go a long way toward interesting research institutions and funding agencies in these sorts of projects.
What else did we learn from all this?
As you may have noticed, the development of this project was in itself an exercise in networking. Only Jon and Rob work in the same geographical location. We all learned what to expect from each other and how to work with one another through mail, electronic mail, phones and meetings. Maintaining deadlines and mileposts was a constant challenge, but in the end, we put together the pieces without killing each other.
Louis_Rossi@uml.edu