Browser-based Large Data Visualization
Modern browsers are becoming a viable platform for application development, including interactive graphical applications. Frameworks like D3 can help to quickly create truly stunning web applications. However, even though an enormous amount of effort has been spent on optimizing browser performance and especially the performance of the Javascript interpreters underlying, there are significant limits when it comes to handling large data, which often means more than 10,000 records already.
The goal of this project was to evaluate the ability of a browser-based environment to interactively handle large amounts of data. The goal here was to be able to interact with data consisting of several million records. As a starting point we chose to use parallel coordinates visualization as it is graphically simple and very flexible when it comes to many dimensions per record.
Our implementation is based on
WebGL to speed up the proces. But in contrast to other WebGL-based visualizations we use WebGL not only for graphics, but to do practically all the calculations, from selecting the parts of the datasets that are affected by the current filters over generating the geometry to the actual drawing.
As a result we are able to interact smoothly with 1 million record datasets. Using parallel coordinates on datasets of this size brings up some interesting new problems in terms of overloading the display with geometry, which can be avoided by using floating-point accumulation, which is only possible through WebGL.
(Images partially blurred for confidentiality protection.)
- Started:
- 2012-08-01
- Ended:
- 2013-05-15
- Past Students:
-
Murali Krishna Pusala
- Funded By:
- NSF Center for Visual and Decision Informatics (CVDI)
