The tariff simulation used in the HPF Virtual Workshop was used as a source of data to be visualized. This DX image shows the historical development of each one of the trading relationships developed from the 100x100 Tariff Case Study dataset. The color of the tube denotes the volume of sales between supplier i and buyer j, where the volume is non-zero.
Each tube indicates the history of that relationship, starting with iteration 1 (in back) and ending with iteration 1000 (in front). The trading has not yet reached equilibrium (there are another 3010 iterations to go) but is nearly there. Trading between any two markets increase when the product can be sold at a price the buyer is willing to pay while still making a profit for the supplier, subject to the tariff policy in effect. The picture clearly shows that once an (i,j) pair drops to zero, it does not start up again; rather, strong relationships get stronger (the tube goes from blue to yellowish red).
More detail concerning the DX program used.
Workload data from the LoadLeveler batch system was the source of data for this visualization. There was a major upgrade in the scheduling system to EASY/LL in April, so the February and March data reflects the old scheduler while the July, August, through October data reflects the new one. Data for these graphs are calculated as follows: Each job is placed in a bin depending on job duration (the back axis, rounded up to the nearest integral hour) and the number of processors used (binned according to the range on the right axis). The user node time (job duration times number of processors used) is then added to the value in the bin.
The height of the glyphs is proportional to the accumulated user node time in each bin. The color reflects the number of jobs in each bin. One can see, for example, that there are may single-processor jobs (red-colored glyphs along the back row), but that they contribute very little user node time. Conversely, there are few long-running, highly parallel jobs, but such jobs account for the vast majority of work done.
One important trend is the emergence of jobs using 257 or more processors after installation of the EASY/LL scheduler. In August, for example, nearly 10% of the total user node time come from such large jobs. Prior to this time none were run, primarily because the old scheduler had a great deal of trouble scheduling them.
Another trend is the shift from 32-way to 64-way as the most common range of processors used. This is due both to the growing ability of users to make use of more processors and to better scheduling of large jobs by EASY/LL.
More detail concerning the DX program used.
A final example will show how many functions of DX are combined. The example program will be general enough to illustrate most of the main functions of the system without mixing in specific scientific data. We choose to use landscape generation as a "data set" to be visualized.
We will generate and visualize a fractal landscape, including the landform, clouds, rivers and trees. The landform is a 2D scalar field, which will be visualized as a surface and contour lines. The surface is pseudocolored by altitude. Clouds are generated by a 3D fractal density field and visualized in two different ways: by treating the density as numerical haze (volume rendering) and by fitting polygons (isosurface extraction) to a certain density level. The gradient of the landscape altitude is calculated as a vector field used to generate "streamlines" which become streams on the landscape. Trees are generated as glyphs and scattered over the landscape.
Thus we cover 2D and 3D scalar fields, 2D and 3D vector fields, isosurfaces, volumetric rendering, streamlines, glyphing, and pseudocolor control without having to teach a specific scientific discipline. Note, however, that all the data for this example is generated in the program. In general, applications you write will need to import data from an outside source.
More detail on the program.
The following link loads DX and runs the program. If you have never used this DX launcher from the Web before be sure to read (footnote.1). You will see an image window and a control panel. By default, the control panel is set to draw the land surface, with water, and some trees. You can turn on streams and clouds. If you turn on streams, you may need to modify the "Stream starts" list to make the streams run at visible locations. Turning on all options, particularly clouds, uses a lot of memory. On smaller machines, you may get "out of memory" errors from DX.
