Sie sind auf Seite 1von 3

A Feasible Method to combat against DDoS Attack

in SDN Network

Nhu-Ngoc Dao1 , Junho Park1 , Minho Park2 , and Sungrae Cho1


1
School of Computer Science and Engineering, Chung-Ang University, Seoul, South Korea
2
School of Electronic Engineering, Soongsil University, Seoul, South Korea
Email: dnngoc@uclab.re.kr, jhpark@uclab.re.kr, mhp@ssu.ac.kr, and srcho@cau.ac.kr
AbstractIn Software Defined Network, the controller is The structure of this paper is organized as follows. Section
so vulnerable to flooding attack. By injecting spoofed request II describes how attackers exploit SDN network vulnerabilities
packets continuously, attackers make a burdensome process to the to take resource consumption attack by flooding technique.
controller, cause bandwidth occupation in the controller-switch Then, we identify the characteristics of DDoS user during
channel, and overload the flow table in switch. The final target attack time. Based on that, we proposed the suitable policies
of attackers is to downgrade or even shutdown the stability and
quality of service of the network. In this paper, we introduce a
for DDoS user and other valid users in section III. The
feasible method to protect the network against Distributed Denial simulation and evaluation will be present in section IV. Finally,
of Service attacks more effectively. section V shows the conclusion and defines our future work.
KeywordsSDN, Openflow, DDoS, DoS
II. H OW TO ATTACK SDN NETWORK BY DD O S?
I. I NTRODUCTION In [4], Shin et al. investigated flooding attacks against SDN
network. The method is described below.
Software defined network (SDN) was born with a great
mission to change the way that existing network architectures The first step is to identify the SDN network. Recall
and devices are still doing, in term of specializing device that, almost traditional networks usually have a pre-configured
operation and network deployment to reach an intelligent forwarding table. Therefore, it needs no additional time to
network. In SDN architecture, the control and data planes are process and create a flow entry for a new incoming packet. In
decoupled, enabling the control of network programmable and contrast, the controller in SDN must take a little time to issue
the underlying infrastructure abstracted for applications and a new flow entry for a new packet. Then, it adds more time to
network services [1]. handle the first packet in comparison with following packets.
Based on this knowledge, attackers can identify whether a
The power of SDN has been being proven day by day,
network is SDN or not by checking the difference between
spreads in various areas, from enterprises to carriers and
respond times of first packet and following packets. If the
service provider environments, from small local area network
difference is greater than a defined threshold, then the network
to public cloud architectures. In most cases, the SDN shows its
is recognized as SDN network.
strong success in providing reliability, effectiveness, simplicity,
flexibility with lower cost [2]. Finally, the last step is just to inject more random fake
However, there are a number of challenges that need to packets into SDN network. The controller must process and
be resolved, especially in security matter. Because of the generate corresponding flow entries harder and harder. The
natural feature of centralized controller, it can easily become a new entries will occupy overall flow table in switch very soon.
potential target of attackers. Whenever a new packet arrives in From that time, the switch cannot serve users normally. The
SDN network, and the switch can not find any matched flow quality of service is downgrade or even shutdown.
entry, it will forward the packet to the controller to ask about
how to handle. It is a good chance for an attacker to deplete III. O UR PROPOSED METHOD
the resources and threaten the network availability. A. Analyzation of user behavior
Recently, there have been proposed solutions developed to
When collecting and analyzing data traffics of the Univer-
decrease the impact of the secuirty problems. Some ideas try to
sity of Auckland network and a Small ISP Trace in one month,
setup specific policies and/or classify the security requirement
Peng et al. [5] realized that around 90% of frequent users sent
data and normal data to separate process. Some of others
at least 5 packets to each destinations. In vice versa, abnormal
customize the framework to make all components co-operate
users transmitted less than 5 packets per connection (error
dependably and securely. However, none of them focus closely
packets or DDoS packets). The minimum number of packets
enough to defence against DoS attack [3].
per connection of a frequent user is denoted by n. In other side,
In this paper, we propose a feasible approach of source- there are around 60% of the IP addresses appeared on only one
based IP filtering technique to defeat DDoS attack. Our method day in the two weeks period. Most of them are infrequent users
works on Openflow protocol, analyzes the user traffic to detect or DDoS attacking addresses. We define the average number
and prevent the attack. of connections which the frequent users establish, is k.

978-1-4799-8342-1/15/$31.00 2015 IEEE 309 ICOIN 2015


Forwarded packet ci The counter of IP address in T table
The minimum number of packets per
n
connection
k The average connections of freq. users
- Set a new entry with
short timeout; s The statistic counter of IP address
- Increase ci by 1;

Y Request statistic
ci k counter s from switch.

Do nothing. Y
sn Block.
Fig. 2. Simulation network topology
N

- Update its entries DDoS attacking user, 01 malicious user and 01 frequent user)
with normal timeout; connected to the SDN network (Fig. 2).
In the switch, the flow table has capacity of 10.000
Fig. 1. The workfow of proposed method
entries. Initially, it is empty. The normal entry has
hard timeout and idle timeout equal to 600 seconds
and 60 seconds. The entry for DDoS attacking user
Therefore, the characteristics of DDoS addresses are to
has hard timeout and idle timeout equal to (60, 10)
initiate less than k connections (in a constant duration) and
seconds.
to generate less than n packets per connection.
The malicious user has IP address 10.0.0.1. It injects
B. The proposed method spoofed packets to the switch infinitely. For each
packet, the destination IP address is generated ran-
We define a temple table (T table) in the controller. The T domly.
table is used to store source IP addresses of forwarded packets
from the switch. Each unique IP address has a counter ci to The DDoS attacking user sends spoofed packets to
track the number of arrived packets. the switch infinitely. For each packet, the source and
destination IP addresses are generated randomly.
During attack time, whenever a new packet forwarded by
the switch arrives at the controller, the controller assumes The frequent user has IP address 10.0.0.2. It es-
that it might be from the DDoS attacking address firstly. The tablishes 5 different connections to the server, and
controller creates a new specific entry with hard timeout and transmits 10 packets per connections.
idle timeout, of which values are smaller than those of normal Assume that we already analyzed the collected data of
entries, to limit its lifetime. Then, the source IP address is frequent users and set the values of k and n to 3 and 7,
updated into T table for tracking, and its counter ci is increased respectively. The network topology is simulated by OPNET
by 1 (Fig. 1). modeler.
When ci reaches k, we need to analyze its data traffic char-
acteristic by requesting the average number of packet counter s. B. Evaluation and discussion
If s is greater than n, this means the source address established
The simulation results are shown below:
and transmited real data connections. In other words, it is
a frequent user. Hence, the controller issues a modification The packets from the malicious user are dropped after
message to reset all hard timeout and idle timeout of its it floods 3 packets. The entry has source address
existing entries to normal value [6]. 10.0.0.1 and action DROP is installed in the flow table.
In vice versa, if s is smaller than n, this is malicious traffic. The DDoS attacking users packets are handled by
The controller dispatches a dropped rule for the address to the corresponding entries with short hard timeout 60s and
switch. idle timeout 10s.
The frequent user is served by normal entries which
IV. S IMULATION AND EVALUATION have hard timeout 600s and idle timeout 60s.
A. Simulation network topology
In the switch, the number of entries is derceased 49%
To simulate the proposed method, we build a small topol- after 100s because of block entries of malicious traffic and
ogy which consists of 01 webserver and 03 PC clients (01 flow entries with short timeouts (Fig. 3). The number of

310
550 20

Bandwidth of the controllerswitch channel (kbps)


500 18
450 w/o proposed method
w/ proposed method 16
Number of entries in flow table

400
14 w/o proposed method
350 w/ proposed method
12
300
10
250

200 8

150 6

100
4
50
2

0 100 200 300 400 500 0 100 200 300 400 500
Time (seconds) Time (seconds)

Fig. 3. Numer of entries in flow table Fig. 5. Bandwidth occupation in the controller-switch channel

4500
V. C ONCLUSION AND FUTURE WORK

4000
SDN is expected to replace the existing traditional network
w/o proposed method
with a lot of advanced features. However, it is facing with
Number of packets arrive at the controller

w/ proposed method
3500 many security challenges. In this paper, we propose a feasible
method to combat against DDoS flooding attack. Although
3000
the method can decrease the impact of DDoS attack, but not
2500 enough when the amount of attack traffic is very huge.

2000
In future work, we need to optimize the method for
adaptive with variable environments and protocols. Implemen-
1500 tation in real SDN network is required for confirmation of its
performance and effect.
1000

500 ACKNOWLEDGMENT
This research was supported by Basic Science Research
0 100 200 300 400 500
Time (seconds) Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Science, ICT & Future
Planning (NRF-2013R1A1A1076105)
Fig. 4. Number of packets arrive at the controller

R EFERENCES
entries remains around 265 entries (includes normal entries [1] https://www.opennetworking.org.
for frequent users, block entries for malicious user, and short [2] S. Sezer, S. Scott-Hayward, P. Chouhan, B. Fraser, D. Lake, J. Finnegan,
N. Viljoen, M. Miller, and N. Rao, Are we ready for SDN? Implemen-
timeouts entries for DDoS attacking users). The bandwidth tation challenges for software-defined networks, IEEE Communications
occupation in controller-switch channel and the total packets Magazine, vol. 51, no. 7, 2013.
arrive at the controller are decreased 51% and 49% respectively [3] S. Scott-Hayward, G. OCallaghan, and S. Sezer, SDN security: a
(Fig. 4 and Fig. 5). survey, in Proc. IEEE SDN4FNS, 2013.
[4] S. Shin and G. Gu, Attacking Software-defined networks: a first fea-
Because we set short values to the hard timeout and sibility study, in Proc. the second ACM SIGCOMM workshop on Hot
idle timeout of all new forwarded packets from the switch to topics in software defined networking, 2013.
the controller, the impact of DDoS attack is downgraded and it [5] T. Peng, C. Leckie, and K. Ramamohanarao, Protection from distributed
denial of service attacks using history-based IP filtering, in Proc. IEEE
improves the protection against the entry overflow attack. After ICC, 2003.
the controller analyzes the data traffic behavior reported from [6] OpenFlow Switch Specification Version 1.3.4, ONF Specifications,
the switch, the malicious source addresses are also cancelled Open Networking Foundation, 2014.
by block entries. However, the frequent users still are treat as
normal.

In order to get more effectiveness, we can tune the values


of parameters k and n corresponding to different network
environments.

311

Das könnte Ihnen auch gefallen