A Study on Network Traffic Rendering by a Hyper-Object

互联网被如此多的人以各种方式使用，并且有越来越多的因素给互联网通信带来意想不到的困难。无论是管理员还是终端用户都希望了解当前网络传播趋势的质的变化。定性方面的可视化呈现有助于直观地理解目前正在网络上发生的复杂现象。我们的研究目的是通过隐喻来抽象网络访问中的特征数据，对网络利用进行可视化的呈现。首先，我们假设了一种情况，比如父母希望看到他们的孩子使用互联网，以避免接触有害信息的风险。本文提出并实现了一种网络访问监控系统，该系统工作在连接Internet的儿童PC机上。看门狗系统通过抓包的方式对互联网的访问进行检查，从下载的网页文本中抽取单词。通过电子邮件定期向家长报告访问趋势分析的图形化表示。我们已经评估了数据包捕获和趋势分析的开销。然后，我们对互联网流量监控进行了假设，并提出了一种基于通信流量动态聚类分析的流量图形化呈现方法——超对象。超对象具有虚拟的三维形状，如球体，聚类结果映射在其上，以便直观地理解网络中的复杂现象。集群是通过自组织映射在虚拟球体上动态形成的一组具有相似流量模式的主机。我们利用实际网络流量数据对原型实现和可视化实验进行了评估。

The Internet is used by so many people in various ways, and there are growing factors to bring unexpected difficulty in the Internet communications. Not only administrators but also end-users want to know qualitative changes of the communication tendency in the current network. Visual presentation for the qualitative aspects is helpful to understand the complex phenomena intuitively that is happening on the network now. Our research objective is to give the visual presentation on the network utilization by a metaphor to abstract the characteristic data in the network access.At first, we have supposed a situation such as parents want to see their children's use of the Internet to avoid risks of contact with harmful information. We have proposed and implemented the Web access watchdog system which works on a children's PC connected to the Internet. The watchdog system always checks the Internet access by packet capturing, and samples the words from downloaded text of Web pages. A graphical presentation of the access tendency analysis is periodically reported to the parents via e-mail. We have evaluated the overhead for packet capturing and tendency analysis.Then, we have supposed the Internet traffic monitoring and proposed a hyper-object which is a graphical traffic rendering approach based on a dynamic clustering analysis of communication traffic. The hyper-object has a virtual 3D shape like a sphere on which the clustering results are mapped to cause intuitive understanding of complex phenomena in the network. A cluster, which is a set of hosts with similar traffic patterns, is dynamically formed on the virtual sphere by self-organization mapping. We have evaluated the prototype implementation and the visualization experiments by using actual network traffic data.