The deployment of a system such as the MBone should be a careful balance between distribution efficiency, resource availability, redundancy in case of failure, and administrative policy. We hope that this series of visualizations and continued work in this area will help ISPs and administrators of campus networks cope with a growing infrastructure by illustrating where optimizations or better redundancies could occur within and across network boundaries. This process has already begun: when our co-author involved with MBone deployment (BF) saw the initial 3D visualization in February 1996, he was inspired to encourage many maintainers of suboptimal tunnels to improve their configuration.
Visualization of the MBone has provided important insight into the existing MBone structure as well as specific problems in the current distribution framework. The 3D visualizations are far more useful for people working in the MBone engineering process than the textual data, and can serve as an educational medium for the general public. However, their interpretation requires a great deal of operational context and they could be misleading for those without a broad understanding of the underlying technologies.
Both the visualization techniques and the geographic mappings that we have developed are valuable tools for the visualization of other network topologies. These tools have also been applied successfully to other domains such as the hierarchical WWW caching system, wide area Internet routing, and congestion measurement. We hope that these visualizations will encourage network providers to make available geographic, topological, and performance information for use in visualizations that could facilitate Internet engineering on large and small scales.
One possible future direction is to create an animation of the evolution of the structure over time using tunnel data archived at regular intervals. Interactive analysis of larger scale topologies, such as the mapping of the MBone onto the underlying unicast framework, or investigating general unicast Internet routing problems, will require a broader set of visualization and data collection techniques. Both general level of detail control and additional layers of abstraction will be critical for interactive manipulation of a structure of this complexity. We would also like to explore other visual metaphors. One strong advantage to our current ``arcs on globe'' geographical layout is that viewers understand it immediately with little explanation. The disadvantage is that tunnels between co-located machines are not shown. A graph embedding where edge length and internode distance are determined by some other data can be a very effective medium for conveying a different understanding of the properties of interest. Such a representation, not tied to the geography of the Earth, is amenable to display in 3D hyperbolic space [9]. However, it can be quite a challenge to obtain the relevant data for many network attributes, for instance delay variation as a metric of goodness.
Another interesting possible use for the tools we have developed is to manage the deployment of the next generation IP protocol, version 6 (IPv6, sometimes referred to as IPNG). Developed over the past two years to address specific engineering concerns related to scalability and packet processing, IPv6 is now approaching standardization, and service providers and network architects are beginning to consider the steps involved in a graceful transition to this new routing and forwarding infrastructure. During the transition tunnels will serve to lash together sections of the new IPv6 Internet, analogous to those that form the MBone today. Appropriate visualization will be extremely valuable in the management of this dynamic Internet-wide migration process.