200 likes | 212 Views
This research paper evaluates the PCAV system, a real-time monitoring system for anomaly-based intrusion detection using visualization. The paper discusses the main idea, algorithm, and evaluation of the PCAV system.
E N D
PCAV: Evaluation of Parallel Coordinates Attack Visualization Hyunsang Choi, Heejo Lee {realchs, heejo}@korea.ac.kr Computer and Communication Security Laboratory Korea University, Korea Joint Workshop between Security Research Labs in Korea and Japan, Kyushu University, Kyushu, Japan, Feb 7 – Feb 9, 2006
Contents 1. Overview 2. Main Idea of PCAV 3. Visualization 4. Algorithm of PCAV 5. Evaluation http://ccs.korea.ac.kr
Introduction Overview Main Idea Algorithm Evaluation Visualization PCAV (Parallel Coordinates Attack Visualization) Propose anomaly-basedreal-time monitoring system with visualization approach Visualization approach Real-time monitoring Anomaly detection Early detection http://ccs.korea.ac.kr
Characteristics: Internet Attacks Overview Main Idea Algorithm Evaluation Visualization <H.Kim et.al.,IEEE Networks 2004> • Large scale Internet attacks • Worm • Source spoofed DDoS attack • Scanning activities • Important Characteristics • One-to-many relationship http://ccs.korea.ac.kr
Selected Parameters Overview Main Idea Algorithm Evaluation Visualization • What we visualize • Selected 4 main parameters in TCP/IP header field IP header TCP header http://ccs.korea.ac.kr
Flow instead of Packet Overview Main Idea Algorithm Evaluation Visualization • Aggregated input data instead of raw traffic Source port Destination port Source IP address Destination IP address ... Data Header Packet Flow Internet http://ccs.korea.ac.kr
Benefits of Visualization B A C D E Overview Main Idea Algorithm Evaluation Visualization Intuitive • Come up with new hypotheses • Deal large noisy • data easily Visualization higher degree of confidence Faster http://ccs.korea.ac.kr
Parallel Coordinates Overview Main Idea Algorithm Evaluation Visualization • How we draw flows on parallel coordinates • Input flow: • Source address • <211.162.35.77> • Destination address • <211.162.35.105> • Destination port • <80> • Average packet length • <1240> 255.255.255.255 255.255.255.255 65000 1500 1240 211.162.35.77 211.162.35.105 80 0.0.0.0 0.0.0.0 0 0 http://ccs.korea.ac.kr c. Host scan d. Port scan Fig. 7. Rescaledattack graphs
Attack Graphs from Real Traffic Overview Main Idea Algorithm Evaluation Visualization 1. Worm Graph - Slammer 2. DDoS attack 3. Hostscan 4. Portscan http://ccs.korea.ac.kr
Attack Signatures Overview Main Idea Algorithm Evaluation Visualization • Graphical signatures and divergences and packet length of implied attack http://ccs.korea.ac.kr
PCAV System Design Overview Main Idea Algorithm Evaluation Visualization • 4 main modules • Sensor • Analyzer • Visualizer • Database • Database • Store flow information – text, image • Remarkably compressed (1/2000) • Replay flows http://ccs.korea.ac.kr
Application Overview Main Idea Algorithm Evaluation Visualization • PCAV 2.0 demo clip http://ccs.korea.ac.kr
Algorithm Overview Main Idea Algorithm Evaluation Visualization • Main algorithm of analyze module http://ccs.korea.ac.kr
Evaluation Overview Main Idea Algorithm Evaluation Visualization • 1Gbps backbone traffic • Windows XP (flow generator), 2003 server (PCAV) • Pentium-4 PC, 1Gbyte memory (about 100MB memory use) http://ccs.korea.ac.kr
Stress Test Overview Main Idea Algorithm Evaluation Visualization • PCAV process 10Gbps trafficwith 98% accuracy. • (Gigabit network exports about 10,000 flows/s) http://ccs.korea.ac.kr
Multiple Attack Overview Main Idea Algorithm Evaluation Visualization http://ccs.korea.ac.kr
False Positive Test Overview Main Idea Algorithm Evaluation Visualization • False positive • Hostscan, DDoS • P2P, web traffic (flash crowd, web crawling), game, chatting (MSN), DNS, mail, streaming, etc • Length filtering effect (flag) • Threshold setting http://ccs.korea.ac.kr
False Negative Test Overview Main Idea Algorithm Evaluation Visualization • False negative • Assumption • Little increased but ignorable • Worm can not be detected without length filtering. • Threshold setting http://ccs.korea.ac.kr
Summary 1 2 3 Main Purpose Early detection Real-time monitoring • Effectiveness • Detect and drawa particular pattern of graph for each attack • Future Work • Auto-threshold configuration • Enhance sampling • process http://ccs.korea.ac.kr
Thanks. Tel: +82-2-3290-3208 Fax: +82-2-953-0771 http://ccs.korea.ac.kr Dept. of Computer Science and Engineering Korea University. Anam-Dong SeoungBuk-Gu, Seoul, KOREA