Tracking dynamic network properties at internet-scale
Shobha Venkataraman · 2012
Many network properties such as security and performance vary considerably in different parts of the Internet, e.g., many measurement studies have shown that malicious activity is more likely to originate from certain regions of the Internet. Such network properties also typically vary over time, e.g., new regions of the Internet may get compromised while previously compromised regions may get cleaned up, or performance degradations may been seen by clients in some regions at certain times of a day. In this talk, we address the problem of discovering regions on the Internet that share a common (and potentially dynamic) network property. We develop online algorithms to leverage the structure inherent in these properties with provable optimality guarantees, and demonstrate that our algorithm achieves high accuracy on large real-world data sets of spam and botnet data.