Improving Network Intrusion Detection with Traffic Denoise
Improving Network IntrusionImproving Network Intrusion
Detection with Traffic DenoiseDetection with Traffic Denoise
Miroslav Stampar (Croatian Government CERT)Miroslav Stampar (Croatian Government CERT)
IntroductionIntroduction
Network Intrusion-Detection Systems (NIDS) generate
lots of false alerts, false alarms, false positives, etc.
Result of thinking that it is better to catch both
suspicious and harmless events than to miss
something
Lots of irrelevant and/or “frustrating” noise
Security managers/technicians are loosing interest
very fast (serious problem)
Detection/noise problemDetection/noise problem
Basis of a NIDS is “detection”
Like metal detector, NIDS doesn't stop network traffic
It highlights potential threats so that the security
guards, like firewalls, can be more effective
Too much noise results with lessened usability and
benefit
As a side effect, increased management cost
Most of all, lack of focus on real adversaries
Real adversaries vs “junk”Real adversaries vs “junk”
What if we could distinguish regular noise from real
adversaries?
Regular noise should be fairly common on Internet
Subjects targeting multiple ranges should not be
considered as big threat as those targeting only us
Various subjects (i.e. “lurkers”) scan 0.0.0.0/0
searching for easy to exploit systems
Unsophistication at its best
Free (skiddie level) vulnerability assessment
How to capture noise?How to capture noise?
It should be visible by different Internet parties
compared to adversaries targeting only our IP range(s)
If unwanted (inbound) traffic is visible from different IP
ranges in predefined time-span (e.g. 1 day) we can
presume that it is a noise
Two ways:
1) Utilizing multiple (public third-party) IP blacklists
2) Constructing own sensor network made of “silent”
boxes
IP blacklists (generally)IP blacklists (generally)
Multiple third-parties publicly share relatively up-to-
date IP blacklists (i.e. IP addresses with “bad
reputation”)
Inbound traffic coming from such IPs could be safely
DROP-ed (in case of active boundary defense – e.g.
IPS)
AlienVault, CINS, Deepviz, VoIPBL, Turris, etc.
IDSes, honeypots, server logs, “threat intel”, etc.
SSH, HTTP, Telnet, VoIP, NTP, DNS, etc.
Denoise (methodology)Denoise (methodology)
Suricata (ELK) IDS on /20 network (4096 IPs)
Fine tuned ruleset based on ETPro (removed all
“paranoid” rules, rules based on “reputation”,
INFO/POLICY/CORPORATE, etc.) with highly adapted
thresholds
Experimental data based on May 2017 events
Denoise results were obtained by manual processing
(for presentation purposes)
Presenting concept (no final product, yet :)
ConclusionConclusion
Ignoring, DROPing or lowering the severity of events
for noisy IPs should help to focus on real adversaries
Testing results look promising
Although currently only a concept, personal wish is
that this presentation will give somebody an idea to
further evolve it
If anything, collection of traffic on silent boxes around
the 0.0.0.0/0 could pinpoint the wandering (noisy) IP
addresses (more than 1000 per day)