150 likes | 313 Views
Design of Bloom Filter Array for Network Anomaly Detection. Author : Jieyan Fan , Dapeng Wu , Kejie Lu , Antonio Nucci Publisher: IEEE GLOBECOM 2006 Presenter: Hsin-Mao Chen Date: 2009/10/21. Outline. Introduction Background Bloom Filter Array Data Structures Algorithm
E N D
Design of Bloom Filter Array for Network Anomaly Detection Author: Jieyan Fan, Dapeng Wu, Kejie Lu, Antonio Nucci Publisher: IEEE GLOBECOM 2006 Presenter: Hsin-Mao Chen Date:2009/10/21
Outline • Introduction • Background • Bloom Filter Array • Data Structures • Algorithm • Random-keyed Hash Function • Analysis • Performance
Introduction • Distributed Denial of Service (DDoS) attacks are the major threats to the Internet. • The TCP-base DDoS attacks using spoofed source IP address are detected in the edge router through two-directionalmatching.
Background • Two-directional(2D) matching A normal TCP flow generated from one end host to another should have a corresponding flow from the other direction.
Background • Bloom Filter Data1 Data2 Data3 K hash function h1() h2() h3() M-bit vector
Bloom Filter Array • The key idea is to use a Bloom filter array to trade off a amount of accuracy, for much less space and time complexity.
Data Structures • A smaller time slot τ, Γ = w × τ, where w is an integer. • Two arrays of bit vectors, {IVi}, {RVi}, i ∈ Zw. • An array of integers {Ci}, i ∈ Zw. • K hash functions, hi(·), i ∈ ZK. Γ …… τ τ τ IV1 RV1 C1 IV2 RV2 C2 IVw RVw Cw h1(), h2()…hk()
Algorithm • Add new unmatched inbound flow. • Inbound packet is stored in at least one RVj. • Inbound packet is stored in IVi%w. Γ …… τ τ τ IV1 RV1 C1 IV2 RV2 C2 IVw RVw Cw h1(), h2()…hk()
Algorithm • The first matched outbound packet comes. • Outbound packet is not contained in RVj%w. • Outbound packet is contained in IVj%w. Γ …… τ τ τ IV1 RV1 C1 IV2 RV2 C2 IVw RVw Cw h1(), h2()…hk()
Random-keyed Hash Function • One kernel hash function with K randomly generated keys. • We can generate as many keys as we want. • The security issue is solved.
Performance • Trace data provided by Auckland University. • <SA,DA,SP,DP> as the signature of the packets. • 2.4G Hz CPU and 1G memory. • There are 78501441 packets in the trace. • The average processing rate is 265000 packets/second.
Performance Number of unmatched SYN packets