ALVIN: a system for visualizing large networks

Visual display of networks can lead to both better understanding and clear presentation of patterns that can often be hidden [3]. However, effectively visualizing large networks has proven to be difficult, due to the limitations of the screen, the complexity of layout algorithms and the limitations of human visual perception. A good layout algorithm (eg. Spring layout) can easily take quadratic time assuming that the network fits in memory. The graph structure of the Web, for instance, is far too large to hold in the memory of most desktops let alone visualize it. To gain insight into the complexity of the problem, consider the graph structure of the Web at the domain level as shown in Figure 1-a. This network is relatively small, having only 224 nodes, but it is still not easy to find any interesting patterns. Colouring the largest Strongly Connected Component (SCC), as is shown in Figure 1-b, singles out some of the domains that are not in the SCC. One such domain, for instance, is Vatican City (va), linked by a large number of domains in the SCC but is not linking back to any domain in the SCC. Even in this graph, it is not easy to see the connectivity structure of many of the domains in the SCC. Scaling up the visualization to a graph of the Web with millions of nodes at the site level or hundreds of millions of nodes at the page level is quite challenging if not impossible. Our proposed alternative in ALVIN 2 is to refrain from visualizing the entire network. At the core of our methods is sampling. We sample the network and only visualize the sample. Even though the network can be quite large, the size of the sample can be adjusted to match the limitations of the visualization environment.