Quick Links
Project Summary
Persistent and exponential growth in global communication, especially relating to activity on the Internet, continues to drive research and innovation in the field of data visualisation. Traditional approaches used for the capture and analysis of network traffic have become superfluous due to the sheer size of traffic datasets. As a result, there is a high demand for network traffic visualisation systems that offer both efficiency and reliably when working with datasets of a large magnitude.
In 2005, Rhodes Master's student, Jean-Pierre van Riel, succeeded in implementing InetVis, a visualisation tool which, at the time, satisfied the demands of visual traffic monitoring. When significant advancements in computer hardware (observed over the five year span between the original InetVis implementation and the currently enhanced re-implementation) were coupled with the constant rise in network traffic, the InetVis tool suffered major performance issues and could no longer achieve acceptable results.
To resurrect the visualisation tool, we enhanced the underlying structure in a variety of ways. A client server model improves memory and processor usage while at the same time, offers flexibility through modularization, thus allowing easy integration and update capabilities. Multi threading optimizes processing and a redesigned graphical user interface exploits the potential of large screen sizes while maintaining full functionality. These enhancements not only revive the InetVis concept, they offer a means to adapt the efficiency of the tool to allow for its survival within a rapidly-expanding network environment.
Project Status: Completed
The re-implementation proved to be highly successful in utilising the capabilities of new technology to represent significantly larger data sets of network traffic. The InetVis tool was tested by J.P van Riel and results showed that the InetVis tool could handle 450,000 packet events before serious performance issues were experienced. The dotNetVis re-implementation was limited to 8,200,000 packets (due to storage limitations imposed by C#'s List capacity) and its performance was satisfactory.
InetVis was tested on a machine with 1 GB of RAM, a 3.0GHz Pentium IV processor and a 256 MB GPU. Before the 450,000 mark was reached, the openGL component of InetVis managed a stable 25 frames per second. dotNetVis, tested on a machine with 3 GB of RAM, a quad core 2.66GHz Intel Q9400 processor and a 256 MB nVidia 7900GT, managed to display 2 million packets at 60 frames per second. Ultimately, at an acceptable 7 frames per second, dotNetVis could display 8.2 million packets (about 6 months worth of network traffic captured on the darknet) while consuming 728MB of RAM. The server component took just over 5 minutes to process and broadcast the packet data to the client and the client's processor idled at around 60% across all three cores while processing the received data. Once stored and indexed, the processor usage dropped to 1% usage for the dotNetVis client process.
While no improvements on the underlying concept were implemented, a host of perfor- mance enhancements were carried out. Optimization of processor usage, modularization of the input and output process and improvements on memory usage coupled with in- creased memory capacity are some of the enhancements that improve on JP van Riels original 3-D scatter plot visualisation tool - InetVis. This research has proved successful in optimizing the InetVis system to handle much larger datasets as well as o er a much more exible architecture on which future extensions can be built.