Stream-wise detection of surreptitious traffic over DNS

Abstract

The Domain Name System (DNS) belongs to crucial services in a computer network. Because of its importance, DNS is usually allowed in security policies. That opens a way to break policies and to transfer data from/to restricted area due to misusage of a DNS infrastructure. This paper is focused on a detection of communication tunnels and other anomalies in a DNS traffic. The proposed detection module is designed to process huge volume of data and to detect anomalies at near real-time. It is based on combination of statistical analysis of several observed features including application layer information. Our aim is a stream-wise processing of huge volume of DNS data from backbone networks. To achieve these objectives with minimal resource consumption, the detection module uses efficient extended data structures. The performance evaluation has shown that the detector is able to process approximately 511 thousand DNS flow records per second. In addition, according to experiments, a tunnel that lasts over 30 seconds can be detected in a minute. During the on-line testing on a real traffic from production network, the module signalized on average over 60 confirmed alerts including DNS tunnels per day.

Unsupervised learning and rule extraction for Domain Name Server tunneling detection

“…The idea is to apply the algorithm over streaming data, that is, during server operation . A new analysis for anomaly detection is thus triggered over a new block of features once a new feature is available.…”

DNS Tunneling Detection Techniques – Classification, and Theoretical Comparison in Case of a Real APT Campaign

“…The last DNS tunneling detection technique presented in this paper combines both traffic and payload inspection [30]. The purpose is to detect tunnels and other anomalies via statistical analysis in huge volume of DNS traffic near real-time.…”

“…From the table, it can clearly be seen that some techniques cover more than one group. Especially, the hybrid detection technique [30] covers both traffic and payload inspection and both statistical and signature-based analysis. All the surveyed machine learning -based detection techniques can also be grouped into statistical analysis.…”

“…The hybrid inspection technique presented in [30] also pays attention to the differences between each DNS request/response, especially to the fields of DNS messages, such as TXT records. However, their detection module analyzes only the domain names that are classified as suspicious in advance.…”

Refined identification of hybrid traffic in DNS tunnels based on regression analysis

“…The original traffic cannot be directly inspected as it becomes a part of the domain name of the DNS packets. There have been several schemes for discovering DNS tunnels [8–12]. Nevertheless, more detailed information about DNS tunneling is still unavailable to the network inspectors.…”