240 likes | 360 Vues
This presentation by Ming-Yuan Hsu explores the development of a software-based Signature Detection System (SDS) on network cards, specifically designed to enhance cybersecurity through effective intrusion detection and prevention. Utilizing the Aho-Corasick algorithm, the architecture aims to provide efficient, real-time packet inspection without extensive payload processing. The evaluation demonstrates its capability in handling realistic data loads, outperforming traditional firewalls in specific scenarios. The discussion includes design insights, the implementation on an Intel IXP1200 network processor, and future improvements.
E N D
Towards Software-Based Signature Detection for Intrusion Prevention on the Network Card Speaker: Ming-Yuan Hsu
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
Information • Authors • H. Bos • Vrije Universiteit, Amsterdam, The Netherlands • herbertb@cs.vu.nl • Kaiming Huang • Xiamen University, Xiamen, China • kmhuang@xmu.edu.cn • Symposium • A. Valdes and D. Zamboni (Eds.): RAID 2005, LNCS 3858, pp. 102–123, 2006. ⒸSpringer-Verlag Berlin Heidelberg 2006 WNS.Lab.Meeting
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
Introduction (1) • Signature detection system (SDS) in software on the network card. • Distributed firewall • Do not implement payload inspection at all. • FPGAs • Complex to modify. • CardGuard is intended to protect • Single end-user’s host • Small set of host connected to a switch WNS.Lab.Meeting
Introduction (2) • Author’s goal has been to make the SDS • An inexpensive device • Competitive with large firewalls. • Fast enough to handle realistic loads. WNS.Lab.Meeting
Distributing the firewall • Most current approaches to IDS/IPS • High-performance firewall/IDS at the edge. • All internal nodes are assumed to be safe. • All external nodes are considered suspect. WNS.Lab.Meeting
Drawbacks (compared to a distributed firewall) • Doesn’t protect internal nodes • Attacks originate within the intranet. • Represent the intranet’s link to the outside world. • The traffic is vary large • Payload scans difficult/ infeasible • Attacker can send a large number harmless packets. • Protect a heterogeneous collection of machines. • Close all ports except a select few. • It’s inconvenience to users. • Configuration is more complex. WNS.Lab.Meeting
The IXP1200 Network Processor • CardGuard implemented • Implemented on an IXP1200 network processor unit. • Employ the Aho-Corasick algorithm WNS.Lab.Meeting
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
Aho-Corasick • A string match algorithm. • Make the dictionary to finite state machine. • A simple example • The dictionary include “aba, abcd” WNS.Lab.Meeting
Inline • In-memory • Store DFAs in Aho-Corasick is a trie WNS.Lab.Meeting
The input “bababcdab” • Aho-Corasick return • aba X 1 • abcd X 1 WNS.Lab.Meeting
Aho-Corasick Example a deterministic finite automaton (DFA) for the Slammer worm identifies 5 different patterns WNS.Lab.Meeting
The signatures of Slammerworm are • h.dllhel32hkernQhounthickChGetTf • hws2 • Qhsockf • toQhsend • Qhsoc • Aho-Corasick could found signatures at: • State 32、35、40、42、50 WNS.Lab.Meeting
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
CardGuard is implement on a Intel IXP1200 NPU board. • Contains 2 Gigabit ports. • Packet reception/transmission • Newer version is IXP2800 • Its efficiency is better than IXP1200. WNS.Lab.Meeting
The ports are used for all data between hosts and the NPU. • CardGuard is designed as a plug-and-play IDS. • To protect a set of hosts connected to a switch. • No reconfiguration of the end-systems is encessary. WNS.Lab.Meeting
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
Evaluation • This is a typical result in Aho-Corasick. • The times of the different level are visited. WNS.Lab.Meeting
First experiment • Use tcpreplay(max rate : about 50Mbps) • Second experiment • Examine the number of cycles. • Packets of various size • Ten state transitions in the DFA. • Single thread could process 52.5Mbps for maximum-sized non-TCP packets. WNS.Lab.Meeting
Final experiment (A stress-test ) • Packets sent by iperf. • Evaluate the throughput under worst cast. • The payload needs to be checked from start to finish. • It’s not a realistic scenario. • Send packet is checked in its entirely. WNS.Lab.Meeting
Outline • Information • Introduction • Aho-Corasick • Architecture • Evaluation • Conclusion WNS.Lab.Meeting
Conclusion • The hardware used in CardGuard is rather old. • The principles remain valid for newer hardware. • CardGuard represent a first step. • Intrusion detection on a NIC in software WNS.Lab.Meeting