Stirring Up a Hive of Trouble
If you ever have to evaluate a network, whether it is a network of IT hardware, individuals or chemical interactions you have many options for visualizing how that network functions. Until recently, however, there was never any clear way to see what that network was doing. Hive plots are just that missing tool, and are something you need to learn to use.
We are all familiar with traditional node-edge or tree diagrams for visualizing the structure of a network and what connects to what. Those tools are great at displaying connectivity and path information for where stuff flows, but those tools alone fall short in 2 important contexts:
- Traditional node-edge diagrams do not show you what is going on inside that structure. You can view the path between any set of nodes, but you don't know where more traffic is versus less, or if a handful of nodes are trafficking the bulk of the load.
- Traditional node-edge diagrams turn into undecipherable hairballs pretty quickly. A network of 10 or 20 nodes is easy wrap your head around, but try to put 500 interconnected nodes on one plot and it's very unlikely you will gain any meaningful information.
What hive plots do is suppress that structural information in return for performance information. You cannot see how a network is structured based on a hive plot, but you can see what is going on. Even more importantly, as the network changes over time the hive plot remains effectively the same. This means that once you find a hive plot that produces meaningful information for your network, you can feel confident that as your network evolves in structure that same hive plot will show you how your structural changes have changes what is going on. So if you're an IT engineer a hive plot can give you a signature of how traffic travels in your system with respect to source, sink and broker nodes. If you change your topology, the hive plot doesn't change so it will tie you an apples-to-apples comparison on how that signature changed based on your structural changes.
The down sided of hive plots is that they aren't easy to think through for a novel situation. As Martin Krzywinski related in a recent email exchange "hive plots are a lot like scatter plots, in that you need to spend time really thinking through what axes matter", but once you do the display is insightful.
Currently I am working on a hive plot to evaluate social network performance. More updates as that work progresses.