Comments on "Using Visualization Techniques to Study Distributed Shared Memory Machines"

Analyzing the performance of distributed shared memory machines is a complex problem and visualization has a big role to play in helping this analysis. The author argues that a generic visualization tool allowing creation of new metaphors is a better alternative than allowing a fixed set of views or quantities. This is an important point, one which is especially relevant in the current context. A related point advocates the use of logical, application-specific boundaries for partitioning data (when required) instead of generic ones. This, while a valid goal, might be rather difficult to acheive in practice. The author suggests inverted plots (in the context of load visualization; the parts of data structures are annotated by the identifier - such as color - of the processor they reside on) as a possible solution. This is a powerful technique and will probably also help in a more general context.

A temperature plot is discussed as a metaphor for communications and network activity. In the plot, the color coded temperature gives the frequency of messages passing through the node. A simple extension to this plot would be to add a second "temperature" giving the frequency of messages originating from or destined for the node. This makes it possible to see whether a node is a hotspot intrinsically (due to the algorithm) or because of routing decisions.

Extension of graphical railway schedules to 3D is a nice idea, but it suffers from the common problem of 3D graphs - they often end up confusing the user and almost always take more time to understand. This point was made by the author himself earlier in the paper.

The temperature plot visualization can be used for visualizing load balance also - this is an alternative to the Kiviat diagrams discussed in the paper. Now communication and loads can be shown together, thereby permitting the study of relationships between the two. If this addition of variables to the plot is continued, the plot converges to the animated glyph representation discussed in Joel Baxter's paper.

Chetan Rai