Graph Layout through the VCG Tool

glish Version to appear with Addison Wesley 22 ple and use heuristics, but they are very fast and enable to explore very large graphs in reasonable time. Further work might address the following: Improve the stability of the layout: Similar graphs must create similar pictures. Allow incrementality of the layout: Adding an edge or a node to an layout should not cause a complete relayout. Allow diierent layout algorithms for nested subgraphs: Each subgraph can be laid out by a specialized variant of a layout algorithm. The combination of such algorithms must be analyzed. Improve the spline drawing routine. The implementation of the tool is based on the diploma thesis of Iris Lemke Lem94] (VCG for SunView). It runs with SunView and X11 on many diierent platforms (SunOS, IRIX, IBM AIX, HP-UX. : : :) and can produce diierent forms of output (PostScript, PBM, PPM). The tool is The times for parsing and rank assignment are obviously independent of the selected layout algorithms. The time for the additional ne tuning phase innuences t r not very much. For trees, the ne tuning phase has no eeect. However, the omission of the ne tuning phase for real graphs leaves more dummy nodes and more edge segments, such that the layout is less readable. It is not clear, whether the barycenter weight or the median weight results in better layouts. For the trees, the quality and speed of both methods is equal: they remove all crossings and result in nearly the same layout. For the graphs, the median weight removes more crossings in 5 cases, while the barycenter weight is better in 6 cases. In 5 cases, crossing reduction by barycentering is considerably faster (t c in g.3 Bft, Bnt, g.4 Bnt, g.6 Bft, g.7 Bft), in 4 other cases, the calculation of median weights is faster (g.2 Mft, Mnt, g.4 Mft, g.5 Mft). As we see for graph 5, graph 6 and graph 7, the time for the calculation of the median weight increases much more depending on the degree of nodes, because all adjacent nodes of v must be sorted in order to calculate the median value of a node v. Thus barycen-tering has advantages if the average degree of nodes is large. Depending on the selected weights for crossing reduction, the times for the further processing of the graphs may diier. Typically, barycentering results in a more symmetrically ordering …