Download presentation
Presentation is loading. Please wait.
Published byBrendan Chase Modified over 9 years ago
1
Botnets Usman Jafarey Including slides from The Zombie Roundup by Cooke, Jahanian, McPherson of the University of Michigan
18
BotSniffer Slides made by Andrew Tjang Paper: BotSniffer: Detecting Botnet Command and Control Channels in Network Traffic by Guofei Gu, Junjie Zhang, and Wenke Lee (NDSS 08)
19
Motivation ● Botnets serious security threats ● Realtime Command+Control from centralized source ● Use characteristics of this Command+Control to detect botnets in sstems
20
Contributions ● Identify characteristics of C&C in Botnets ● Capture spatial-temporal correlation of network traffic to detect botnets ● Implement anomaly based detection algorithms as Snort plugins ● Evaluation of BotSniffer on real world traces ● Show botnets can be detected with high accuracy and low false positive rate
21
Command & Control ● Centralized control of bots in botnets ● Can be push (i.e. IRC) or pull (i.e. HTTP) ● Difficult to detect because protocol usage similar to normal traffic, low traffic volume, few bots, encryption
22
Spatial-Temporal correlation ● Invariants to all botnets 1. need to connect to central server to get commands 2. respond to commands ● perform tasks and report back (keeping long connection, or making frequent connections) ● Responses: message/activity response ● Multiple bots in channel likely to respond in similar fashion Leverage “response crowd” Bots have stronger/consistent synchronization and correlation in responses than humans do.
23
BotSniffer Architecture ● Monitor Engine Examines network traffic, detects activity response behavior, suspicious C&C protocols ● Correlation Engine Group analysis of spatial-temporal correlation, similarity of activity or message responses connected to same IRC/HTTP server
26
BotSniffer Architecture Illustrated
27
Group Analysis ● Intuition: P(botnet | 100 clients send similar messages) > P(botnet | 10 clients send similar messages) IF botnet, THEN more clients more likely to form homogeneous cluster IF not botnet, THEN unlikely to send similar messages
28
Evaluation ● Datasets University wide network IRC traffic 2005-2007 (189 days) All network wide traffic (10min/1-5h) Botnet traces (synthetic) ● Honeypot (8hr) ● IRC server logs ● Modified bot software in virtual environment ● Implemented 2 botnets using HTTP
29
Results – Normal Trace
30
Detection
31
Attacks on Botsniffer and Their Defenses ● Misuse of whitelist Whitelists not necessary Can use soft whitelists ● Encryption Doesn’t affect activity response ● Long & random response delays ? ● Random noise packets Activity response unaffected
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.