The driving force behind our work is the desire to solve a real problem: understanding the geography and topology of the MBone in order to keep it from breaking. Since repair will require the cooperation of many MBone users and system administrators around the world, a fundamental goal of the project is to make the interactive 3D maps available to as many people as possible.
In pursuit of this objective, we chose to use the VRML 1.0 (Virtual Reality Modeling Language) format for distributing 3D data with embedded hyperlinks across the World Wide Web . VRML popularity is growing quickly, and there are already many 3D browsers available for a wide variety of platforms . By making our 3D maps available on the Web as VRML files, we allow the MBone community to interactively explore the MBone structure. We believe that disseminating 3D data files will lead to a clearer understanding of the problems and possible solutions than merely distributing still pictures or even videos.
We thus use a VRML 3D visualization system to display the tunnels as arcs on a globe. The VRML support for hypertext allows us link the graphical representation of each tunnel to a textual description. The user can determine the specific administrative endpoints that correspond to a given tunnel by clicking on the tunnel in the 3D viewer. The link points to hypertext that contains the IP address, hostname, location (city/state/country), and latitude and longitude of the tunnel endpoints.
The visualization engine used for most of our development has been done on an adaption of the WebOOGL 3D Web browser , built on top of the Geomview platform . Most VRML browsers do not support interactive changes of color, linewidth, and position on a per-group basis. In fact, the VRML 1.0 file format does not facilitate per-group interactions (this restriction is will be lifted in future VRML 2.0 systems). The WebOOGL/Geomview system uses a different Web-enabled file format internally, which we convert into VRML. We use the Geomview-based visualization system to dynamically change values until we arrive at a useful configuration. We can then feed these values back into the grouping and visualization construction phase to create distributable VRML files. The Geomview-based system also visually highlights the selected tunnel in the 3D scene through color and linewidth changes, in addition to showing the hypertext information in another window.
We use a globe constructed by Stuart Levy of the Geometry Center using the CIA World Map database. We draw outlines of the continents on a solid-colored sphere rather than using a texture map of the real Earth to reduce both computational requirements and extraneous visual clutter. Most of our development work is on SGI platforms. The VRML files that we disseminate use a lower resolution version of the outlines, so that lower-end systems can maintain reasonable frame rates during interaction.
We construct the arcs on the globe using the hostname/latlong for the endpoints of each tunnel in the database described above, and compute the shortest geodesic arc between those lat/longs on the Earth's sphere. Using the endpoints of the spherical geodesic as the endpoints of the arc, we loft the arc above the surface of the sphere. The height of the arc depends on the length of the geodesic, using essentially the same equation as the SeeNet3D system . We briefly experimented with using the same height for all arcs, but the display quickly became very cluttered. Having the arc height depend on its length lends visual emphasis to long arcs, an advantage for our application since we want inordinately long tunnels to stand out. Likewise, short tunnels are least obvious, which is appropriate since such tunnels impact global congestion the least. We also impose a minimum arc height requirement so that even very short tunnels remain visible.
Although the geographic representation of the data is vastly easier to comprehend than the raw textual form, the current MBone structure is sufficiently complex that further visualization techniques are desirable. We made a similar set of visualizations using lines on a two dimensional projected map of the earth, and found that the lines representing the tunnels were far too dense to allow for interpretation. The move to the globe-based three dimensional representation gave the display much more locality and enhanced the users ability to selectively display tunnels. The user can rotate, translate and zoom to focus on an area of interest. We have found that the capability to change the center of rotation from the center of the earth to a user-chosen point on its surface to see a ``horizon view'' of the tunnels (as in Figure 2), is particularly helpful to understand local structure of the shorter tunnels.
We have implemented thresholding in order to reduce the visual clutter. Figure 3 shows a thresholded view where we clip the tunnel drawing to a radius around the endpoints. Once again, this is particularly useful when focusing on local rather than global structure.
Finally, grouping allows us to visually distinguish between categories of tunnels. We use a regular-expression based filter which allows us to categorize based on any text attribute of the original data. For example, in Figure 2 we group based only on the ``info'' field of the tunnels. We use a more powerful filter in Figure 7 which categorizes across multiple fields, in this case both the source and destination hostnames. We can visually distinguish these groups by color and linewidth, and in sufficiently flexible 3D viewers can move a group in relation to the other 3D data. While such motion of course dislocates the tunnels from their geographic reference points on the globe, which would be very disorienting for a single tunnel, it can be very useful when comparing tunnel groupings complex enough that the tunnels alone connote geographic structure.