Database System Support for Data Visualization

一张图片胜过千言万语。用单一的图像取代复杂的叙事可能已经成为新闻业的陈词滥调，但这是数据探索的绝对要求。毕竟，一个人可以读一千个单词，但不可能看到，更不用说理解十亿个数据点了。可视化为理解数据提供了不可或缺的工具。数据可视化的过程可以从高层次的角度描述为由两个步骤组成：处理数据，然后对其应用一种可视化算法。在过去的50年里，人们在数据可视化方面做了大量的工作。有趣的是，大部分工作都集中在可视化过程的第二步。这是可以理解的，因为数据集相对较小，可视化工具的性能通常由图形算法的效率决定。随着大数据集的出现，数据可视化工具的设计必须发生根本性的转变。数据检索和处理时间现在主导了可视化过程的性能，因此不能忽视。没有有效的数据库支持，可视化，特别是交互式可视化是不可能的。我们提出的研究计划就是要解决这个问题。我们将研究针对不同类型数据的不同可视化技术，并分析数据处理引擎的性能，该引擎对要可视化的数据进行预处理。我们将提供新的数据处理技术，以提高后端系统的性能，并将其与前端可视化步骤集成。我们期望建立这种系统的原型，并通过实验证明它们的可行性。

One picture is worth a thousand words. The idea of replacing a complex narrative with a single image may have become a cliché in journalism, but it is an absolute requirement in data exploration. After all, one may be able to read a thousand words, but cannot possibly look at, let alone understand, a billion data points. Visualization offers an indispensable tool in understanding the data. The process of data visualization can be described from a high-level perspective as consisting of two steps: processing the data, then applying one of the visualization algorithms to it. There has been significant work done on data visualization over the last 50 years. Interestingly, most of this work concentrated on the second step of the visualization process. This was understandable, as data sets were relatively small and the performance of the visualization tools was often determined by the efficiency of the graphics algorithms. With the advent of large data sets a radical shift in the design of data visualization tools has to be made. Data retrieval and processing time now dominates the performance of the visualization process, so cannot be ignored. Without efficient database support, visualization, in particular, interactive visualization is impossible. The research program we propose addresses this very problem. We will look into different visualization techniques over different types of data and analyze the performance of the data processing engine that pre-processes data to be visualized. We will offer new data processing techniques to improve the performance of the back-end system and integrate them with the front-end visualization step. We expect to build prototypes of such systems and demonstrate experimentally their viability.