INVESTIGATION OF COOPERATIVE DEFENSE
AGAINST DDOS
Igor Kotenko and Alexander Ulanov
St. Petersburg Institute for Informatics and Automation, 39, 14 Liniya, St.-Petersburg, 199178, Russia
Keywords: Denial of Service and other attacks, DDoS Defense, Simulation of Internet Attacks and Defense.
Abstract: The paper considers a new approach and a simulation environment which have been developed for
comprehensive investigation of Internet Distributed Denial of Service attacks and defense. The main
peculiarities of the approach and environment are as follows: agent-oriented framework to attack and
defense investigation, packet-based simulation, and capability to add new attacks and defense methods and
analyze them. The main components of the simulation environment are specified. Using the approach
suggested and the environment implemented we evaluate and compare several cooperative defense
mechanisms against DDoS (DefCOM, COSSACK, and our own mechanism based on full cooperation). The
testing methodology for defense investigation is described, and the results of experiments are presented.
1 INTRODUCTION
One of the most dangerous classes of the Internet
attacks is Distributed Denial of Service (DDoS). The
prospective DDoS defense system is supposed to
work due to the cooperation of various network and
global defense mechanisms functioning both in local
networks and in the whole Internet.
The distributed cooperative DDoS defense
mechanisms are the ones that implement the defense
due to resource roaming (e.g. Server Roaming
(Sangpachatanaruk, etc., 2004)), change of resources
quantity, resource differentiation (e.g. Market-based
Service Quality Differentiation (Mankins, etc.,
2001), Transport-aware IP router architecture
(Wang, etc., 2003)), authentication (e.g. Secure
Overlay Services (Keromytis, etc., 2002)) and also
the mechanisms that implement traceback (e.g.
Gateway-based mechanism [Xuan, etc., 2001]) with
packet marking or signature storing, including
pushback, auxiliary packet generation, etc.
There are various defense methods with resource
differentiation that use rate-limiting. They are in
dedication of different traffic volumes for different
protocols and lowering the load of defense system or
target system (to allow them implement the
countermeasures effectively). Applying the rate-
limiting can be also the result of attack traceback
mechanisms. Agents-limiters are distributed in the
defended or ISP subnet and implement rate-limiting
according to the given protocols. Such methods are
mostly represented in cooperative defense
mechanisms.
Our goal is to develop the simulation
environment which can help investigate the Internet
attacks and defense mechanisms and elaborate well-
grounded recommendations on the choice of
efficient defense mechanisms. In the paper we
investigate the aspects of component cooperation for
COSSACK (Papadopoulos, etc., 2003), DefCOM
(Mirkovic, etc., 2005) and our own full cooperation
approach as well as trying to develop a new
approach to the investigation of cooperative defense
mechanisms based on agent-oriented packet-based
simulation.
The work is organized as follows. Section 2
outlines suggested approach for simulation.
Section 3 defines the DDoS defense mechanisms
investigated. Section 4 describes the simulation
environment developed. Section 5 presents the
testing methodology for defense mechanisms
investigation. Section 6 analyzes the results of
experiments fulfilled. Conclusion surveys main
work results and future research.
2 SIMULATION APPROACH
It is suggested to represent the investigated
processes as an interaction of various teams of
180
Kotenko I. and Ulanov A. (2007).
INVESTIGATION OF COOPERATIVE DEFENSE AGAINST DDOS.
In Proceedings of the Second International Conference on Security and Cryptography, pages 180-183
DOI: 10.5220/0002128801800183
Copyright
c
SciTePress
software agents in the dynamical environment
defined on the basis of the Internet model (Kotenko,
Ulanov, 2006). Aggregated system behavior is
shown in local interactions of particular agents.
There are at least three different classes of agent
teams: teams of agents-malefactors, teams of
defense agents, teams of agents-users. Agents of
different teams can be in indifference ratio,
cooperate or compete up till explicit counteraction.
Agents of attack teams are divided, at least, into
two classes: “daemons” that realize the attack and
“master” that coordinates other system components.
The class of attack is defined by the following
parameters: a packet sending intensity and an IP-
address spoofing technique (no spoofing, constant,
random, random with real IP addresses).
According to the general DDoS defense
approach suggested the defense agents are classified
into the following classes: information processing
(“sampler”); attack detection (“detector”); filtering
(“filter”); investigation (“investigator”). Samplers
collect and process network data for anomaly and
misuse detection. Detector coordinates the team and
correlates data from samplers. Filters are responsible
for traffic filtering using the rules provided by
detector. Investigator tries to defeat attack agents.
Defense team jointly implements certain
investigated defense mechanism.
Defense teams can interact using various
schemes. Moreover, a new class of defense agent –
“limiter” – is introduced. It is intended for the
implementation of cooperative DDoS defense. Its
local goal is to limit the traffic according to the team
goal. It lowers the traffic to the attack target and
allows other agents to counteract the attack more
effective.
There are three types of limiting: by the IP-
address of attack target; by the IP-addresses of
attack sources; according to the packet marking.
Detector sets limiting mode using detection data.
3 DEFENSE MODELS
The main attention in cooperative mechanisms is
given to the methods of distributed filtering and
rate-limiting. These methods help to trace the attack
sources and drop the malicious traffic as far from
attack target as possible.
DefCOM (Mirkovic, etc., 2005) works in the
following way. When “Alert generator” detects the
attack it sends the attack messages to the other
agents. “Rate limiter” agents will start to limit the
traffic destined to the attack target. “Classifier”
agents will start to classify and drop the attack
packets and to mark legitimate packets.
DefCOM is simulated as follows. “Alert
generator” agent is based on “detector”, “Rate
limiter” – on “limiter” agent, agent “Classifier” – on
“filter”.
COSSACK (Papadopoulos, etc., 2003)
consists two main agent classes: “snort” and
“watchdog”. “Snort” (IDS) prepares the statistics on
the transmitted packets for different traffic flows; the
flows are grouped by the address prefix. If one of the
flows exceeds the given threshold then its signature
is transmitted to “watchdog”. “Watchdog” receives
traffic data from “snort” and applies the filtering
rules on the routers. Agent “snort” is based on the
agent “sampler”, “watchdog” – on the agent
“detector”. It makes the decision about attack due to
data from “snort”. Agent “filter” is used to simulate
filtering on routers.
COSSACK cooperation is in the following: when
“watchdog” detects the attack it composes the attack
signature and sends it to the other known
“watchdogs”. “Watchdogs” try to trace in their
subnets the attack agents that send attack packets;
when they detect them the countermeasures are
applied.
Proposed approach. There are used the
following four classes of defense team agents:
“samplers”, “detectors”; “filters”; “investigators”.
Agent teams are able to interact using various
cooperation schemes: no cooperation; filter-level
cooperation; sampler-level cooperation; poor
cooperation; full cooperation. The main aspect of
full cooperation is that team which network is under
attack can receive traffic data from the samplers of
other teams and apply the filtering rules on filters of
other teams.
Figure 1 shows the full cooperation defense
system configuration proposed by the authors.
Figure 1: Proposed defense system configuration.
INVESTIGATION OF COOPERATIVE DEFENSE AGAINST DDOS
181
4 SIMULATION ENVIRONMENT
The simulation environment architecture consists of
the following components (Kotenko, Ulanov, 2006):
Simulation Framework, Internet Simulation
Framework, Multi-agent Simulation Framework,
Subject Domain Library.
Simulation framework is a discrete event
simulator. Other components are expansions or
models for Simulation Framework. Internet
Simulation Framework is a modular simulation suite
with the realistic simulation of Internet nodes and
protocols. Multi-agent Simulation Framework
allows realizing agent-based simulation. Subject
Domain Library is a library used for imitation of
processes from subject domain and containing
modules that extend functionality of IP-host:
filtering table and packet analyzer.
This architecture was implemented for multi-
agent simulation of DDoS attack and defense
mechanisms with the use of OMNeT++ INET
Framework and software models developed in C++.
Agent models implemented in Multi-agent
Simulation Framework are represented with generic
agent, attack and defense agents. Subject Domain
Library contains various models of hosts, e.g.
attacking host, firewall, etc., and also the application
models (attack and defense mechanisms, packet
analyzer, filtering table).
5 EXPERIMENTS
The developed simulation environment is used to
investigate cooperative defense mechanisms. We are
investigating the methods effectiveness, their
functioning in various cooperative schemes and in
different networks with different attacks.
The following cooperation schemes are
investigated: DefCOM; COSSACK; full cooperation.
The following main admission for the simulation
was made. Each cooperative defense mechanism
(COSSACK, DefCOM or our approach) is based on
its own attack detection method. We proposed to use
for investigation of cooperative defense the same
methods, e.g. Hop counts Filtering (HCF), Source IP
address monitoring (SIPM), Bit Per Second (BPS),
etc. The use of the same detection methods allows
investigating various cooperative mechanisms in
equal conditions.
The investigation is supposed to be done on the
basis of analysis of the following main parameters:
the amount of incoming traffic before and after filter
of team which network is the attack victim; false
positive and false negative rates of the defense team
which network is under attack.
6 RESULTS OF EXPERIMENTS
Figures 2-4 show the traffic inside (line of squares)
and before (other lines – of dots, rhombs and
crosses) the attacked subnet for the DefCOM,
COSSAK and full cooperation schemas accordingly.
The Figure 2 consists of four graphs since the traffic
was measured at the entrance to the subnet.
Attack starts at 300 seconds. The random real IP
spoofing technique is applied as the most
complicated for detection (the addresses for
spoofing are taken from the same network). SIPM is
used as the defense method. The router is placed
there, it has four interfaces. The significant traffic
increase is noticed in the beginning of attack
(Figures 2-4). But in the area of 350 seconds the
defense system detects the attack and traffic is being
limited before the defended subnet ad being filtered
in the source subnets (Figures 2-4).
DefCOM’s rate limiter proceeds to limit the
traffic because of the high attack traffic volume
(Figure 2). But this cooperation schema succeeds to
keep the traffic on the acceptable level due to
limiting and to applying filtering rules. In
COSSACK the filtering rules are applied and the
attack signature is sent to the other defense
components. In full cooperation the attack signatures
are sent to other cooperating teams which apply the
filtering rules in their subnets. The traffic on the
entrance to the defended subnet is decreased due to
their actions.
Figure 2: Traffic inside and before the attacked subnet for
the DefCOM schema.
Experiments showed the effectiveness of various
cooperative DDoS defense. The best cooperative
schema is the full cooperation. Sampler cooperation
played the key role in the defense. DefCOM schema
shows the stable containment of attack traffic due to
a limiter on the entrance to the defense subnet and
SECRYPT 2007 - International Conference on Security and Cryptography
182
classifiers in the source subnets. COSSACK schema
has the similar traffic level in the defense subnet, but
outside it the traffic stays enough high.
Figure 3: Traffic inside and before the attacked subnet for
the COSSACK schema.
Figure 4: Traffic inside and before the attacked subnet for
the full cooperation.
7 CONCLUSION
The main results of the paper consist in
implementing the simulation environment developed
by the authors for packet-level agent based
simulation of various cooperative defense schemas
against DDoS (DefCOM, COSSACK, and our own
based on full cooperation). The goal of the paper
was to evaluate the effectiveness of these schemas
and compare them.
The multitude of experiments we implemented
demonstrated that full cooperation shows the best
results on blocking the attack traffic. It uses several
defense teams with cooperation on the level of filters
and samplers. DefCOM advantage is in using a rate
limiter before the defended network. It allows
lowering the traffic during attack and letting the
defended system work properly. COSSACK is one
of the examples of peer-to-peer defense network. It
uses attack signatures transmission between agents
to apply the filtering rules near the source.
Future work is concerned with improving the
functionality of the simulation environment and
investigating new cooperative defense mechanisms.
ACKNOWLEDGEMENTS
This research is being supported by grant of Russian
Foundation of Basic Research (Project No. 07-01-
00547), program of fundamental research of the
Department for Informational Technologies and
Computation Systems of the Russian Academy of
Sciences (contract No 3.2/03), Russian Science
Support Foundation and partly funded by the EU as
part of the POSITIF project (contract No. IST-2002-
002314) and RE-TRUST (contract No. 021186-2).
REFERENCES
Keromytis, A., Misra, V., Rubenstein, D., 2002. SOS:
Secure Overlay Services. In Proceedings of ACM
SIGCOMM'02, Pittsburgh, PA.
Kotenko, I., Ulanov A., 2006. Simulation of Internet
DDoS Attacks and Defense. In Proc. of ISC 2006.
Samos, Greece. LNCS, Vol. 4176.
Mankins, D., Krishnan, R., Boyd, C., Zao, J., Frentz, M.,
2001. Mitigating Distributed Denial of Service Attacks
with Dynamic Resource Pricing. In Proceedings of the
17th Annual Computer Security Applications
Conference. ACSAC’01.
Mirkovic, J., Robinson, M., Reiher, P., Oikonomou, G.,
2005. Distributed Defense Against DDOS Attacks. In
University of Delaware CIS Department Technical
Report CIS-TR-2005-02.
Papadopoulos, C., Lindell, R., Mehringer, I.,
Hussain, A., Govindan, R. 2003. Cossack:
Coordinated suppression of simultaneous attacks. In
Proceedings of DISCEX III.
Sangpachatanaruk, C., Khattab, S.M., Znati, T., Melhem,
R., Mosse, D., 2004. Design and Analysis of a
Replicated Elusive Server Scheme for Mitigating
Denial of Service Attacks. In Journal of Systems and
Software, Vol.73(1).
Wang, H., Shin, K.G., 2003. Transport-aware IP Routers:
A Built-in Protection Mechanism to Counter DDoS
Attacks. In IEEE Transactions on Parallel and
Distributed Systems, Vol.14(9).
Xuan, D., Bettati, R., Zhao, W., 2001. A Gateway-Based
Defense System for Distributed DoS Attacks in High
Speed Networks. In Proceedings of the 2nd IEEE
SMC Information Assurance Workshop, West Point,
NY.
INVESTIGATION OF COOPERATIVE DEFENSE AGAINST DDOS
183