DagSemProc.09211.2.pdf
- Filesize: 88 kB
- 4 pages
The seamless operation of the Internet requires being able to monitor and visualize the actual behaviour of the network. Today, IP network operators usually collect network flow statistics from critical points of their network infrastructure. Flows aggregate packets that share common properties. Flow records are stored and analyzed to extract accounting information and increasingly to identify and isolate network problems or security incidents. While network problems or attacks significantly changing traffic patterns are relatively easy to identify, it tends to be much more challenging to identify creeping changes or attacks and faults that manifest themselves only by very careful analysis of initially seemingly unrelated traffic pattern and their changes. There are currently no deployable good solutions and research in this area is just starting. In addition, the large volume of flow data on high capacity networks and exchange points requires to move to probabilistic sampling techniques, which require new analysis techniques to calculate and also visualize the uncertainty attached to data sets.
Feedback for Dagstuhl Publishing