Visualizing complex structures using Graphviz

The Graphviz [1] program has long been well-established in the open source landscape (see the "Out of Bell Labs" box). In many cases, this tool does its job unnoticed in the background, for example, automatically generating images on a web server from a database via a script.

The AsciiDoc [3] markup language has the ability to embed Graphviz data directly. The compiler automatically interprets the code sections into corresponding graphs.

Plugins are available for Doxygen [4] and many wiki platforms. Graphviz also has a presence in Puppet [5], and it is used for creating resource graphs.

To prepare directed and undirected graphs (see also the "Graph Types" box), the command-line tools dot , neato , fdp , circo , and twopi can help. These are all part of the Graphviz package, and they use specific algorithms from graph theory to reduce edge lengths, separate subgraphs, and recognize connectivity components [6].

Workflow

You first use a text editor to describe the elements and their interdependencies and save this as a text file.

The next step is translating the data with one of the aforementioned tools into an output format (e.g., GIF, PNG, SVG, PDF, or Postscript). The following command creates a PNG file from graph.dot :

$ dot -Tpng graph.dot -o graph.png

Building the dot file is quite easy. First, you define a graph (Listing 1, line 1). The keyword digraph creates a directed graph. Always remember the curly bracket after the keyword.

01 digraph G {
02 A -> B [style=dotted];
03 A -> C [color=red, label="A Label"];
04 C [shape=box, style=filled, color="0.7 0.7 1.0"];
05 B -> C;
06 }

In the lines that follow, you specify the three nodes A, B, and C with corresponding attributes to create the nodes and edges. Each line ends with a semicolon. For the node name, you can use either an identifier (as in the C programming language), a number, or a quoted string.

Dot recognizes blank, control, and other punctuation characters as well. Multiple lines are marked with \n for a line break. The label then appears in the middle of the code.

The -> creates an edge between two points in the direction indicated. The curly brackets on line 6 end the definition statement. You control the graph's appearance by passing attributes, which are in square brackets. These attributes can be global or local and appear on lines 2 and 3.

Line 2 defines a dotted line, and line 3 defines a red line with a label for the edge with the label attribute followed by an equal sign and the label content in quotes. In line 4, you define a box for node C, which is colored blue on the inside. The exact color value comes from the RGB components. Figure 1 shows the source code and the result in an image viewer.

Figure 1: A color visualization of a directed graph with three nodes.

With Graphviz you can, among other things, visualize directory hierarchies. An undirected graph can show a tree structure where possibly undesired folders are located. With the folder option, you can assign small icons to nodes very much like those you see for file manager objects (Listing 2).

01 graph {
02 "/" [shape=folder];
03 "/boot" [shape=folder];
04 "/usr" [shape=folder];
05 "/home" [shape=folder];
06 "/var" [shape=folder];
07 "/frank" [shape=folder];
08 "/peter" [shape=folder];
09 "/" -- "/boot";
10 "/" -- "/usr";
11 "/" -- "/home";
12 "/" -- "/var";
13 "/home" -- "/frank";
14 "/home" -- "/peter";
15 }

You translate this file very much like the earlier example. Figure 2 shows the resulting graph. If you use one of the other tools, it will translate the description without an error, but it might represent the folders with circles, which is contrary to the usual view of a filesystem.

Figure 2: Graphviz comes with a little icon that you can use to represent folders in a directory hierarchy.

With a bit of skill, you can generate the dot files from a script. In this way, you could automatically get an overview of all the branches in your filesystem.

Binary Trees

Trees exist as data structure made up of nodes and edges, where the root is the top node. Binary tree have two kinds of nodes – outside (without children) and inside (with exactly two children). The primary application of such binary trees is structuring and creating the most efficient search for data sets.

Graphviz takes the burden off of creating these structures if you use the program in the right way. Listing 3 shows how you can create such a binary tree with the software. After setting all nodes as rectangles on line 2, you then define the structure and content of the nodes in lines 4 through 8.

01 digraph G {
02 node [shape=record, height=0.1];
03
04 node0 [label = "<l> | <m> H | <r>"];
05 node1 [label = "<l> | <m> D | <r>"];
06 node2 [label = "<l> | <m> A | <r>"];
07 node3 [label = "<l> | <m> P | <r>"];
08 node4 [label = "<l> | <m> W | <r>"];
09
10 node0:l -> node1:m;
11 node1:l -> node2:m;
12 node0:r -> node3:m;
13 node1:r -> node4:m;
14 }

Each node has three elements, a reference to the left node (labeled l ), a middle element (labeled m ) with content, and the right node (labeled r ). These elements you will later refer to by name.

In line 10, you connect left node0 with middle node1, and so forth in the lines that follow. You separate the node name with that of the matching element with a colon, which determines where the arrows point. The follow-up representation creates the desired boxes and lines (Figure 3).

Figure 3: Graphviz automatically puts the arrows for the binary search tree in the right places if you use the correct syntax.