CC Data Visualization Library: Visualizing large amounts of scientific data through interactive graph widgets on ordinary workstations

Abstract

A big problem today is that there are currently few or no solutions available for workstations that are able to provide both good interactivity and fast response times while visualizing large amounts of data. Using existing solutions often cause crashes or freezes due to the assumption that all data will fit in main memory and can be processed as such. The purpose of this thesis was thus to simplify analysis of large amounts of scientific data by creating a small modular and extensible cross-platform graphics library, intended to run on ordinary workstations, capable of handling such data and present it through highly interactive plot graph widgets. Focus was on basic functionality for two-dimensional graphs, namely scatter plot graphs and time series graphs. The created library implements a memory manager that avoids keeping too much data in memory by controlling and partially storing and loading data to and from a temporary file and the main memory as needed. A clipmap-like structure is used to display the graph, and a cache of clipmap image tiles is used in combination with progressive updates to improve performance and responsiveness. Insufficient planning forced some of the functionality and requirements to be dropped, which left room for much improvement, as well as a lot of possible out of scope additions and extensions, such as better utilization of the cache, supporting groups of selected data points, logarithmic scaling, and improvements to the memory manager, renderer and GUI. Comparing the results of a simple review and tests of some existing data visualization libraries it was found that the used renderer had average performance, although it had problems using very large or many clipmap image tiles. The resulting library may not be faster than many existing libraries, but it allows faster interaction with good tile configurations and stands out in that it also allows interaction during loading of data and graph updates, while normally avoid consuming all system resources. Its potential is promising, as it thus allows for faster extraction of interesting portions of the data, giving it many possible applications within several areas.