772 Dezhi Han et al.
DDoS attack is early-warned based on abnormal changes of source IP and destination IP
information entropy of network data stream.
3.1. Traffic Information Entropy Feature
Entropy is an indicator of diversity and uniformity of the microscopic state which reflects
the probability distribution of the system in the microscopic state. It can be seen from
the perspective of communication that random interference in a system is unavoidable.
Therefore, statistical methods can be adopted to describe characteristics of the commu-
nication system. To be specific, take the information source as a collection of random
events whose probability of occurrence is similar to uncertainty in the microscopic state
in thermodynamics; Calculating probability of occurrence in each information source in
the information system to simulate the uncertainty of the system in thermodynamics, thus
forming information entropy [12]. Information entropy has similar meaning to entropy in
thermodynamics and it is an uncertainty indicator of the information system, which may
indicate the amount of information in an information system.
Based on the network traffic information, entropy is defined as shown in Equation (1).
H(X) = E[− log p
i
]= −
X
n
i=1
p
i
log p
i
(1)
In above Equation (1), X represents an information source symbol which has n values:
X
1
...X
i
...X
n
, each value corresponding probabilities are: P
1
...P
i
...P
n
, since each source
symbol appears independent of each other, so there comes to the equation:
X
n
i=1
p
i
= 1 (2)
When DDoS attacks are launched, hundreds of bottled machines will send large streams
of data packets to the target and the attacker, in order to hide its position, will randomly
produce fake source IP addresses for the attacking packets or adopt more advanced reply
flood DDoS attacks. In this case, the amount of requests for source IP addresses moni-
tored by the server will drastically increases and the distribution will be more dispersed.
Moreover, there will be a large amount of request flow flocking into certain service ports
at the server side, and at the same time, the requests distribution for destination IP ad-
dresses which monitored by the server and the destination ports will become concentrated
increasingly. When it occurs to the DDoS attacking, the information entropy of destina-
tion IP and source IP of the data flow that arrived the attacked server, which can reflect
the uncertainty of system by calculating information entropy of destination IP and source
IP, that also can be used for the DDoS attack warning in large-scale network traffic.
Fig. 1 and Fig. 2 are shown as the experimental and test conditions of the public server
for the authors school network center. In the beginning of the first 100 seconds test time,
the public servers to be tested will be attacked by traffic DDoS 30GB, which are issued
by multiple clients in the laboratory. From the detecting results of the gateway to connect
the public server, DDoS attack flow occurred in 100th seconds and it is detected by the
system that the information entropy based on the destination IP and source IP occurs
significant changes. The information entropy based on destination IP decreases rapidly,
while the information entropy based on source IP increases rapidly. The result may certify
that when the information entropy can better reflect the DDoS attack, the server receives