A Blind Detection Method for Tracing the Real Source of DDoS Attack Packets by
Cluster Matching
Yonghong Chen, Xin Chen, Hui Tian, Tian Wang, Yiqiao Cai
College of Computer Science and Technology
Huaqiao University
Xiamen, China
e-mail: {djandcyh, xin_tren}@163.com
Abstract—With the rapid growth of the Internet, the impact of
attacks becomes more serious. IP spoofing makes hosts hard to
defend against DDoS attacks. In this paper, we propose a blind
detection method for tracing the real source of DDoS attack
packets. Tracing the real source of a single-packet is difficult,
so we trace-back a cluster of similar packets rather than a
single-packet by cluster matching. We choose K-harmonic
means clustering method to preprocess the packets according
to our proposed quantitative model, at the same time, we
propose an approach to determine the best number of clusters.
In addition, we propose a novel detection algorithm about
cluster matching for tracing the real source of packet clusters
based on K-harmonic means and our improved silhouette.
Experimental results show that our method can detect the real
source of packets with up to 92.54% accuracy.
Keywords-Distributed Denialof Service (DDoS); traceback;
cluster matching; K-harmonic means; silhouette
I. INTRODUCTION
Distributed Denial of Service (DDoS) attacks are
launched by sophisticated attackers, with a huge amount of
information and congestion, which rapidly exhaust available
resources of target systems and intentionally disrupt network
services. The source IP address in a packet can be spoofed
when an attacker wants to hide himself from tracing.
Therefore, IP traceback is a significance part of defending
against DDoS attacks.
The existing methods for tracing the real source of DDoS
attack packets need packet marking [1], [2] or large amount
of extra packets [3], [4]. These schemes are not suitable for
tracing the real source of packets in the network which do
not use the router with the corresponding configuration.
In this paper, we propose a blind detection method for
tracing the real source of DDoS attack packets. First of all,
we propose using K-harmonic means [5] clustering method
to preprocess the packets. We choose K-harmonic means
clustering method according to our proposed quantitative
model, which could estimate the independence between the
clustering result and the centroid initial position. The
independence of preprocess results is critical in our
algorithm, because we need to do the cluster matching for
trace-back. In addition, we propose an approach to determine
the best number of clusters. Finally, we propose a novel
detection algorithm about cluster matching for tracing the
real source of packet clusters based on K-harmonic means
and our improved silhouette. Experimental results show that
the detection accuracy of our method is up to 92.54%, which
is higher than the previous method [6]. In this paper,
accuracy means correct rate.
The organization of this paper is as follows: We briefly
introduce background knowledge about K-harmonic means
and silhouette [7], at the same time, describe our proposed
algorithm and its difference in Section II. Section III presents
the experimental results and discussion. Section IV
concludes the paper.
II. T
RACE-BACK ALGORITHM
Tracing the real source of a single-packet is difficult, so
we trace-back a cluster of similar packets rather than a
single-packet by cluster matching.
A. The Whole Trace-Back Algorithm
Before cluster matching we preprocess the packets by K-
harmonic means clustering method [5], which cluster the
packets of each sent stream and the received stream into
clusters respectively (See B), at the same time, streams of
DDoS attack would be clustered. To trace the real source of a
received packet, noted as , we do cluster matching as Table
I.
TABLE I. THE PROCESS OF TRACE-BACK ALGORITHM
Step 1
We find out the cluster, noted as
, the packet belongs to
in the received stream.
Step 2
Compute the distance between cluster
and each cluster in
sent streams respectively (See C).
Step 3
Find the cluster, noted as
, in sent streams, that has the
minimum distance to the cluster
.
Step 4
Judge the real source of the packet according to
. That
is to say, if
belongs to the sent stream that are sent from
, the real source of packet is
.
B. Preprocess the Packets by Clustering
In order to make preparation for cluster matching, we
preprocess the packets of each sent stream and the received
stream into clusters respectively. The target of preprocessing
the received stream is to cluster the received packets, which
are sent from the same source, together
1) The degree of independence between the clustering
result and the centroid initial position.
Since we need to do the cluster matching based on the
clustering result, the degree of independence between the
551
978-1-5090-1781-2/16/ $31.00 ©2016 IEEE
2016 8th IEEE International Conference on Communication S oftw are and N etw ork s