Abstract

Discovering the causes of incorrect behavior in large networks is often difficult. This difficulty is compounded when some machines in the network are compromised, since these compromised machines may use deception or tamper with data to frustrate forensic analysis. Recently proposed forensic tools enable administrators to learn the causes of some system states in a partially compromised network, but these tools are inherently unable to (1) observe covert communication between compromised nodes or (2) detect attempts to exfiltrate sensitive data.

In this paper, we observe that the emergence of Software-Defined Networking (SDN) offers interesting new opportunities for network forensics. We sketch the design of an SDN-based forensic system that can be used to investigate a wide variety of faults in data center networks, including previously unobservable attacks such as data exfiltration and collusion between compromised nodes. Our key insight is that the network itself can be used as a point of observation, which gives us a holistic view of network activity. We show that a collection of lightweight middleboxes would be sufficient to support this functionality, and we discuss several additional challenges and opportunities for SDN-based forensic tools.