基于Spark的大数据热图可视化方法

摘要

针对普通客户端浏览和分析大数据困难的问题，结合 Spark 和 LOD 技术，以热图为例提出一种面向大数据可视化技术框架。首先利用Spark平台分层并以瓦片为单位并行计算，然后将结果分布式存储在HDFS上，最后通过web 服务器应用 Ajax 技术结合地理信息提供各种时空分析服务。文中重点解决了数据点位置和地图之间的映射，以及由于并行计算导致的热图瓦片之间边缘偏差这2个问题。实验结果表明，该方法将数据交互操作与数据绘制和计算任务分离，为浏览器端大数据可视化提供了一个新的思路。%It is important to provide data analysts with effective and efficient exploratory tools via web browsers. However, due to the characteristics of big data, current data visualization approaches can hardly display the whole datasets. This paper proposes a general-purpose visualization framework based on Spark and LOD. Firstly, we implement a tile-based parallel computing algorithm for layered datasets with Spark. Secondly, we store the tem-porary results on HDFS. Finally, with Ajax and geographic information, we provide all kinds of spatial-temporal analysis services via web. This paper resolves two problems: one is the mapping of data points from heat map to geographic map; the other is the correction of the marginal error in overlapping areas caused by parallel comput-ing. The experiment results suggest that by separating the data display and manipulation from the data rendering and computing tasks, our method provides a new way for big data visualization via web browsers.