skitter viz using hypviewer
skitter and hypviewer
Recently I spent a couple of days porting the hypviewer library to Solaris 2.7 and FreeBSD 3.1. Hypviewer is a library for viewing large graphs. It uses a layout in hyperbolic space and projects the space into a unit sphere. Hypviewer was developed by Tamara Munzer at Stanford University.I've written a quick program to convert skitter data (in ARTS format) to the format used by the hypviewer file reader. Here are some initial results viewing a large skitter dataset for source riesling.caida.org, with roughly 29,000 destinations. This dataset is from February 18th, 1999.
Note that links are reddish in color on their way out of a node and blue on their way in to a node.
|
This image shows that many paths from our source (riesling.caida.org) traverse 204.70.1.197. The source is not visible, but is in one of the clouds in the left of the sphere. Note the outdegree of 204.70.1.197; it has many next hops. There are also a number of large networks behind this router, as seen by the clusters of nodes and links in the right side of the sphere. |
|
One of the most interesting things about hypviewer is its ability to reduce clutter in the neighborhood of a node (rendering a readable local topology in a huge graph), while still allowing easy navigation of the entire graph. This is due in large part to the use of hyperbolic space, but also the layout algorithm. This particular image shows the neighborhood of 198.32.130.15 while indicating the large portions of the network elsewhere (in the top and left of the sphere). |
|
|
|
|
|
Here's a case where there are many paths through a router (209.83.160.130), but most of the paths appear to be to directly-connected hosts. |
It should be noted that all of the screenshots above were taken from a single run on a single dataset. It's difficult to appreciate this software until you play with it and start realizing its potential. Not so much as a visualization system in and of itself, but as a powerful navigation system for network measrement data. There are interfaces for node and link selection, node searching and animated navigation, grouping, etc. which could prove useful when tied to a user interface for large skitter datasets.
Daniel W. McRobb
CAIDA