Map of the Month
Mappa.Mundi Magazine
Martin Dodge is a Researcher in the Centre for Advanced Spatial Analysis (CASA), University College London and is the creator of the Atlas of Cyberspaces.

» Previous Issues
Atlas of Cyberspace
Buy the Book Today!

Further Reading:

» For more information about this subject, the following resources are recommended.

» (1) The CAIDA homepage have somewhat of a convention of naming their Internet analysis tools after sea creatures. Also see the Walrus graph visualization tool homepage.

» (2) See the Jellyfish pages of the National Aquarium in Baltimore for background information.

» (3) Skitter is one of CAIDA's key projects and comprises a set of tools that are undertaking a large scale, active monitoring of the global Internet from 16 stations in the US, Asia and Europe. Data gathered is enabling analysis of both changing Internet topology and performance.

For more details see the skitter homepage and also their technical paper, Internet Tomography, by K Claffy, Tracie E. Monk and Daniel McRobb, Nature Web Matters, January 7th 1999.

» (4) For background information on focus+context distortion see the Nonlinear Magnification Home Page by Alan Keahey. Perhaps the best known example is Tree Studio from Inxight Software, a nice example of a 2D hyperbolic distortion for use as in interactive website maps.

» (5) For more details see Munzner's Ph.D. dissertation, Interactive Visualization of Large Graphs and Networks, Stanford University, June 2000.

» (6) See CAIDA. As well as recent technical papers, “CAIDA: Visualizing the Internet,” by KC Claffy, Internet Computing, January / February 2001, Vol. 5, No. 1. “Internet measurement and data analysis: topology, workload, performance and routing statistics,” by KC Claffy, NAE, 1999 workshop.

» (7) See MAPNET and Plankton

By Martin Dodge, CASA Map of the Month Archives »

What Does the Internet Look Like, Jellyfish Perhaps?

Exploring a visualization of the Internet by Young Hyun of CAIDA

      The Internet is often likened to an organic entity and this analogy seems particularly appropriate in the light of some striking new visualizations of the complex mesh of Internet pathways. The images are results of a new graph visualization tool, code-named Walrus, being developed by researcher, Young Hyun, at the Cooperative Association for Internet Data Analysis (CAIDA) [1]. Although Walrus is still in early days of development, I think these preliminary results are some of the most intriguing and evocative images of the Internet's structure that we have seen in last year or two.

      A few years back I spent an enjoyable afternoon at the Monterey Bay Aquarium and I particularly remember a stunning exhibit of jellyfish, which were illuminated with UV light to show their incredibly delicate organic structures, gently pulsing in tanks of inky black water. Jellyfish are some of the strangest, alien, and yet most beautiful, living creatures [2]. Having looked at the Walrus images I began to wonder, perhaps the backbone networks of the Internet look like jellyfish?

Click for larger image
An example visualization of Internet topology produced by Walrus
(Courtesy of Young Hyun, CAIDA)

      The image above is a screengrab of a Walrus visualization of a huge graph. The graph data in this particular example depicts Internet topology, as measured by CAIDA's skitter monitor [3] based in London, showing 535,000-odd Internet nodes and over 600,000 links. The nodes, represented by the yellow dots, are a large sample of computers from across the whole range of Internet addresses.

Olive Left Top Corner Spacer Image
What Is CAIDA?
Spacer Image
Olive Right Top Corner

CAIDA, the Cooperative Association for Internet Data Analysis, started in 1997 and is based in the San Diego Supercomputer Center. CAIDA is led by KC Claffy along with a staff of serious Net techie researchers and grad students, and they are one of the worlds leading teams of academic researchers studying how the Internet works [6]. Their mission is “to provide a neutral framework for promoting greater cooperation in developing and deploying Internet measurement, analysis, and visualization tools that will support engineering and maintaining a robust, scaleable global Internet infrastructure.”

In addition to the Walrus visualization tool and the skitter monitoring system which we have touched on here, CAIDA has many other interesting projects mapping the infrastructure and operations of the global Internet. Two of my particular favorite visualization projects developed at CAIDA are MAPNET and Plankton [7]. MAPNET provides a useful interactive tool for mapping ISP backbones onto real-world geography. You can select from a range of commercial and research backbones and compare their topology of links overlaid on the same map. (The major problem with MAPNET is that is based on static database of ISP backbones links, which has unfortunately become obsolete over time.) Plankton, developed by CAIDA researchers Bradley Huffaker and Jaeyeon Jung, is an interactive tool for visualizing the topology and traffic on the global hierarchy of Web caches.

Khaki Left Bottom Corner
Khaki Right Bottom Corner

      Walrus is an interactive visualization tool that allows the analyst to view massive graphs from any position. The graph is projected inside a 3D sphere using a special kind of space based hyperbolic geometry. This is a non-Euclidean space, which has useful distorting properties of making elements at the center of the display much larger than those on the periphery. You interact with the graph in Walrus by selecting a node of interest, which is smoothly moved into the center of the display, and that region of the graph becomes greatly enlarged, enabling you to focus on the fine detail. Yet the rest of the graph remains visible, providing valuable context of the overall structure. (There are some animations available on the website showing Walrus graphs being moved, which give some sense of what this is like.) Hyperbolic space projection is commonly know as “focus+context” in the field of information visualization and has been used to display all kinds of data that can be represented as large graphs in either two and three dimensions [4]. It can be thought of as a moveable fish-eye lens. The Walrus visualization tool draws much from the hyperbolic research by Tamara Munzner [5] as part of her PhD at Stanford. (Map of the Month examined some of Munzner's work from 1996 in an earlier article, Internet Arcs Around The Globe.)

      Walrus is being developed as a general-purpose visualization tool able to cope with massive directed graphs, in the order of a million nodes. Providing useful and interactively useable visualization of such large volumes of graph data is a tough challenge and is particularly apposite to the task of mapping of Internet backbone infrastructures. In a recent email Map of the Month asked Walrus developer Young Hyun what had been the hardest part of the project thus far. “The greatest difficulty was in determining precisely what Walrus should be about,” said Hyun. Crucially “... we had to face the question of what it means to visualize a large graph. It would defeat the aim of a visualization to overload a user with the large volume of data that is likely to be associated with a large graph.” I think the preliminary results available show that Walrus is heading in right direction tackling these challenges.

      However, Hyun points out that it is still early days and over the next six months or so Walrus will be extended to include core functions beyond just visualizing raw topology graphs. For CAIDA, it is important to see performance measurements associated with the links; as Hyun notes, “you can imagine how important this is to our visualizations, given that we are almost never interested in the mere topology of a network.” Walrus has not revealed much new scientific knowledge of the Internet thus far, although Hyun commented that the current visualization of topology “did make it easy to see the degree to which the network is in tangles how some nodes form large clusters and how they are seemingly interconnected in random ways.” This random connectedness is perhaps what gives the Internet its organic look and feel.

      Of course this is not real shape of the Internet. One must always be wary when viewing and interpreting any particular graph visualization as much of the final “look and feel” results from subjective decisions of the analyst, rather than inherent in the underlying phenomena. As Hyun pointed out, “... the organic quality of the images derives almost entirely from the particular combination of the layout algorithm used and hyperbolic distortion.“ There is no inherently “natural” shape when visualizing massive data, such as the topology of the global Internet, in an abstract space. Somewhat like a jellyfish, maybe?

 Copyright © 1999-2001
ISSN: 1530-3314

contact | about | site map | home T-O