CREDIT BASED METHODOLOGY TO DETECT AND DISCRIMINATE DDOS ATTACK FROM FLASH CROWD IN A CLOUD COMPUTING ENVIRONMENT

International Journal of Network Security & Its Applications (IJNSA), Vol.5, No.5, September 2013
DOI : 10.5121/ijnsa.2013.5511 129
CREDIT BASED METHODOLOGY TO DETECT AND
DISCRIMINATE DDOS ATTACK FROM FLASH
CROWD IN A CLOUD COMPUTING ENVIRONMENT
N.Jeyanthi, Hena Shabeeb and Mogankumar P.C.
School of Information Technology and Engineering
VIT University, Vellore – 632 014, Tamilnadu, India
ABSTRACT
The latest trend in the field of computing is the migration of organizations and offloading the tasks to
cloud. The security concerns hinder the widespread acceptance of cloud. Of various, the DDoS in cloud is
found to be the most dangerous. Various approaches are there to defend DDoS in cloud, but have lots of
pitfalls. This paper proposes a new reputation-based framework for mitigating the DDoS in cloud by
classifying the users into three categories as well-reputed, reputed and ill-reputed based on credits. The
fact that attack is fired by malicious programs installed by the attackers in the compromised systems and
they exhibit similar characteristics used for discriminating the DDoS traffic from flash crowds. Credits of
clients who show signs of similarity are decremented. This reduces the computational and storage
overhead. This proposed method is expected to take the edge off DDoS in a cloud environment and ensures
full security to cloud resources. CloudSim simulation results also proved that the deployment of this
approach improved the resource utilization with reduced cost.
KEYWORDS
Cloud, DDoS attack, Flash crowds, Reputation-based, credits.
1. INTRODUCTION
Cloud computing is a subscription-based promising technology which provides everything to its
dependents as ‘service’ on demand basis. It extends the IT capabilities with a viable option of
computation through Internet. As it is a ‘pay-as-u-go’ model, the companies are migrating
towards cloud at a much faster pace. It has also unchained the users from the burden of resource
management and maintenance, which is all done by the Cloud Service Provider (CSP). Users can
access the services from anywhere at any time, provided they have any Internet enabled system
(which can be desktops, laptops, mobile phones, tablets etc) with any browsing software.
The new paradigm, which is an amalgam of various prior models such as distributive computing,
grid computing and utility computing, also encompasses on the techniques like pooling, sharing
and virtualization of resources. Metered service (bill on use) and elasticity (scale up and down on
demand) are other hallmarks of the new model. The cloud comes up with three deployment
models (public, private, and hybrid) and delivery models (Software as a Service [SaaS], Platform
as Service [PaaS] and Infrastructure as a Service [IaaS]). The public cloud is open for all whereas
the access to a private cloud is restricted to the owners and customers of an organization. The
cloud SaaS freed the users/organizations from installing the software they need in their PCs. E.g.
CRM software. The PaaS provides the platform such as programming language that is needed by
users to develop Apps. The network bandwidth, database storage and all are coming under IaaS.

130
Unfortunately, cloud with its rich asset of resources and large number of customers, has obviously
engrossed the attackers also. It has to encounter all the security threats that any other Internet
enabled service do. The list goes like trustworthiness of the CSP, access control, authentication
and identification, availability, policy integration, audit and so on. Among these, the threat against
availably, Distributed Denial of Service (DDoS) attacks, which floods the CSP with illegitimate
traffic is the most challenging, damaging and significant one. As compared to single-tenant
infrastructures, the impact of DDoS on cloud is perilous. Even though cloud has the feature of
rapid resource provisioning, the elastic nature of cloud serves the illegitimate traffic also. This
may lead to exhaustion of critical resources. Again, the organizations which are reliant on these
CSPs may lose millions of dollars due to unavailability of service at correct times. This may even
force them to move on to other CSPs which in turn affect the reputation as well as the income of
CSPs. Thus, this ill-guarded security threat has to be dealt with soon, so as to make use of cloud
benefits to full extend.
Although, various researches are going on in this regard and many solutions exist, there are many
pitfalls for them like computational and communicational overhead, high memory consumption,
cost, usage of critical cloud resources itself for discriminating the attack traffic etc. Most of the
methods allows the attack traffic to arrive at CSP and then only takes actions against mitigation.
In this paper, we are proposing a new framework to defend against the intractable attack of DDoS
by giving reputation to users according to the credits they attained. The observation that attack
traffic exhibits similar flow characteristics is deployed here. The credits of clients who show
similarity in traffic flow are decremented and such requests are dropped. The ill-reputed clients
without any credits are blacklisted and blocked. Our scheme claims less computational overhead
and faster detection of attack. The method treats the well-reputed clients with equal priority and
also presents a notification mechanism to aid the well-reputed users get rid of probable viral
attacks due to which they send request contributing to DDoS traffic.
Remaining of this paper is organized as follows: Section 2 presents the literature survey. Section
3 gives light to the proposed solution and section 4 concludes the paper.
2. LITERATURE SURVEY
The security concerns of cloud environment are the most widely discussed topic today. Various
defense mechanisms exist today for detecting and mitigating the number one threat to cloud
computing, DDoS attack. The elastic cloud can’t distinguish the attack traffic and legitimate
traffic by its own. So, the traffic has to first authenticated and filtered. These include the
approaches like passwords, cryptography, puzzles, trust-based, reputation-based etc.
An overlay based crediting mechanism called OverCourt, has been proposed in [3] in which the
users are classified as well-behaving and ill-behaving based on their behavior. VIP paths are used
for tunneling the requests from well-behaving users and non-VIP path for others. Based on
whether a user gets response from server or not, they are classified. The credits are incremented
when the user gets a response. This method has the advantage that it is an overlay a based
network and needn’t have to modify any existing infrastructure. An overcourt gateway and two
crediting routers do the task of detecting the malicious traffic and discarding them. A credit
decaying mechanism is employed to address the issue of dynamic address allocation. But, the
method assumes that the legitimate users will back off during attack period and this is not always
the case. The criterion for crediting the users based on response is also not a good deal every time
as there are chances that attackers may also get the responses. In [1] the reputation of a flow is
found based on the credit acquired due to its diversity in packet size. The attackers usually prefer
to send small size packets. The flow having LOW reputation is malicious and those falls in HIGH
range of the scale is legitimate. But packet size cannot be considered as a measure of legitimacy.

131
The fine grained capabilities are used in [2] to grant tickets to clients. The clients request for
service is preceded by a ticket request. This request contains the credits and penalty values
acquired by clients in previous interactions. But, this method fails if the attacker is human. Also, a
user can fool the provider by turning hostile after acquiring the ticket.
Apart from these, the information distance [4], inter arrival time of packets, flow correlation
coefficient [5] are various other methods proposed in this regard. The [6] classifies the packets
based on their predictability of arrival rates. None of this method can be considered as an
infallible means of defense against the threat of DDoS attack.
Comparison among the existing Trust based techniques is presented in Table.1.
3. PROPOSED SOLUTION
The proposed solution classifies the clients as well-reputed, reputed and ill reputed based on the
credit values they acquire as a consequence of their behavior. Credit can have value ranging from
LVALUE to HVALUE (which can be fixed by CSP). Clients having credit values greater than a
prefixed value, PVALUE is categorized as well-reputed clients. The requests from such clients
will be tunneled through a special channel, where they can access all critical resources of CSP
with equal priority. The clients who has credits values between the LVALUE and PVALUE is
treated as reputed and such clients are allowed limited access to CSP resources and they are not
tunneled through the special channel. The requests from other clients whose credit value is less
than the LVALUE are dropped and such clients are blacklisted as ill-reputed clients, so that they
are blocked in future also.
Table 1: Comparison Table on the surviving Trust based techniques
Trust based
techniques
Advantages Disadvantages Simulation/
Experiment work
Trust Ticket
Deployment [5]
Simple method
No third party involved in
issuing trust ticket
Data owner can have control
over offloaded data as well
as users
Trust ticket reduces the
interactions between users
and CSP
Data is encrypted twice,
once by data owner, and
next by CSP
Clear logical sequence of
tasks.
Data owner can update the
expiry time of trust ticket of
any user at any time
Overall computation time is
reduced
Limited for SaaS
CSP can insist the users to
share the Key.
Can’t rely on the trust of
Online registration
process
Java Network
Programming-
Emulated Cloud
Environment using
VMware ESXi 4.1
Hypervisor based
platform
Service
Trustiness and
Resource
Legitimacy in
Cloud
Computing [3]
Support for dynamic nature
of cloud
Not experimentally proved. Not proved
experimentally
Above the Trust
and Security in
Cloud
Computing: A
Notion towards
Innovation [4]
Secure channel is
established
Better than SSL
Trustworthiness of CSP can
be questioned
KDC should also be a
trusted entity
Not proven Experimentally
Not proved
experimentally
Use Trust
Management
Module to
Achieve
safeguard both the
customers and providers
cross-cloud environment
Improved flexibility and
Trustworthiness of the so-
called familiar CSPs
Simulation

132
Effective
Security
Mechanisms in
Cloud
Environment
[8]
portability of cloud system.
help to increase the
interoperability
A Novel Cloud
Bursting
Brokerage and
Aggregation
Algorithm for
Multi Cloud
Environment [7]
Interconnectivity
Security
resource sharing
mapping
No experimental results. Not proved
experimentally.
Security Agents:
A Mobile Agent
based Trust
Model for
Cloud
Computing [6]
Mobile agents-load
balancing, fault tolerance,
network management etc.
Attacks on VMs can be
prevented
Data audit & event logging
Session key management
No experimental results
Not proven
experimentally.
A Model for
User Trust in
Cloud
Computing [1]
Evaluated various
parameters influencing the
trust
Online Survey,
Likert Scale
A Trust
Management
Model to
enhance
security of
Cloud
Computing
Environments
[9]
robust, fault tolerant and
secure cloud computing
detect malicious middle
nodes
CloudSim toolkit
Cloud security
using FPGA [2]
Can’t get user data even if
the attacker knows user
credentials.
Security system at user side
Hardware failure of FPGA
Physical access to FPGA
vicinity
Not proven
experimentally
3.1. Assumption
The attacker will usually instigate DDoS attack by finding vulnerable systems in the network
(E.g. having no anti-virus protection) and install dreadful programs in them which can make those
systems to send requests to the target upon the command of the attacker. These vulnerable
systems (zombies) which are distributed across the network sends request packets of similar
pattern to the target as per the instructions in the installed programs. Hence, the attack flow will
be almost similar in nature compared to the flash crowd flow coming from legitimate users. This
observation is used here to distinguish DDoS attack traffic from flash crowd.
Fig. 1 depicts the flow diagram of the proposed credit based concept. When the CSP receives
requests for service from the clients, it is checked whether the resources are getting flooded or
not. In normal case, the system will check whether the user who has sent the request is a new user
or not. If he is a new user, his credit is set to MVALUE and assigned the default path where he
has limited access of CSP’s resources. If the user is an already existing one, his credits are
incremented. In resource overload period, the flow is analyzed to find the similarity and similar
flows are discarded. The credits of senders of such flow are checked. If he is a well-reputed user,
he is notified about the likelihood of presence of some harmful programs in his system. If he is
ill-reputed user, he is blacklisted and blocked from sending request in future. The clients who
contributed to dissimilar flow are considered as legitimate users and their credits are incremented.
They may be allowed with restricted or full access to CSP’s resources based on their credits. They
reach the CSP through the assigned path.

Figure 1.
3.2. Crediting Mechanism
Initially, all clients are assigned a credit value, MVALUE, which is the mean of LVALUE and
HVALUE as represented in equation (1),
Under normal circumstances, the credits of all clients are incremented according the following
equation:
where is an increment factor that can be fixed randomly by the CSP.
Under attack, the CSP will experience resource overload and the credits of
almost similar request are reduced.The credit values of such clients are decremented according to
following equation:
where is a decrement factor fixed by the CSP
If those clients were already in the well
or Trojan attack and the decrement in the credit. Thus, they can take necessary actions to come
out of viral attack and escape from being penalized further. The credits of other clients are
incremented as per eqn. (1). Traffic from such clients is considered to be as flash crowds and is
processed for providing the requested service.
A credit expiring mechanism is employed which gradually decrements (until MVALUE) the
credit value acquired by a client with time
1. Working model of the credit based system
HVALUE as represented in equation (1),
.
normal circumstances, the credits of all clients are incremented according the following
is an increment factor that can be fixed randomly by the CSP.
Under attack, the CSP will experience resource overload and the credits of clients who send
is a decrement factor fixed by the CSP.
If those clients were already in the well-reputed list, they are notified about the chance of a virus
qn. (1). Traffic from such clients is considered to be as flash crowds and is
processed for providing the requested service.
credit value acquired by a client with time to address the issue of dynamic IP address allocation.
133
normal circumstances, the credits of all clients are incremented according the following
clients who send
they are notified about the chance of a virus
qn. (1). Traffic from such clients is considered to be as flash crowds and is
to address the issue of dynamic IP address allocation.

3.3. Credit Based Architecture
As shown in Fig.2, the proposed architecture consists of a forward proxy server, which acts as a
gateway between the user’s private network and CSP network, a load balancer and a coordinator
router. The traffic to datacenters is routed through respective f
which can analyze the flows’ time of receipt, route, and rate of flow. The flow routers will
communicate the time of receipt and rate of flow to the coordinator router and coordinator router
compares the results it got from all flow routers to distinguish the traffic from attacker and
legitimate users.
The requests from the users are received by the proxy server, which finds whether there is any
resource overload in the CSP. If there is no resource overload, the proxy se
whether the requests are coming from new user or not. After assigning the credit MVALUE for
new user, their requests will be forwarded to the CSP through default path. For others, credits are
incremented and well reputed users’ requests are
reputed users are assigned the default path.
Figure
When flooding occurs, the proxy server notifies the load balancer and the
distributed to the to the flow routers which are not busy at that instant. The information regarding
the state (busy or not) of flowrouters will be communicated to the load balancer by the
coordinator router. Flow router finds time of recei
flow, the coordinator router informs the proxy server about the legitimate clients. The credits of
such clients are incremented by the proxy server and their reputation is checked based on which
they are assigned path to the datacenters in cloud.
3.3.1. Virtues of Proposed Concept
• The method doesn’t have to maintain any predefined profiles of traffic, or history of
communication. Only thing that has to store is the credit and corresponding reputation of
each client.
• The credit expiry mechanism doesn’t allow the credits acquired by one client to be
inherited by anyone due to dynamic IP address allocation.
Special path
Default path
Architecture
router. The traffic to datacenters is routed through respective flow routers (R1, R2, R3, R4 & R5),
om all flow routers to distinguish the traffic from attacker and
resource overload in the CSP. If there is no resource overload, the proxy server will check
incremented and well reputed users’ requests are tunneled through special path whereas the
reputed users are assigned the default path.
ure 2. Architecture of Credit based method
When flooding occurs, the proxy server notifies the load balancer and the traffic will be
coordinator router. Flow router finds time of receipt and flow rate. After discarding the suspicious
ssigned path to the datacenters in cloud.
of Proposed Concept
The method doesn’t have to maintain any predefined profiles of traffic, or history of
The credit expiry mechanism doesn’t allow the credits acquired by one client to be
due to dynamic IP address allocation.
Special path
Default path
134
low routers (R1, R2, R3, R4 & R5),
om all flow routers to distinguish the traffic from attacker and
rver will check
tunneled through special path whereas the
traffic will be
pt and flow rate. After discarding the suspicious
The method doesn’t have to maintain any predefined profiles of traffic, or history of
The credit expiry mechanism doesn’t allow the credits acquired by one client to be

• As flow routers do the function of flow analysis and load balancer distributes the tasks to
these routers which are not busy at the instant, there won’t be any flooding.
• Notification mechanism
malicious programs.
3.3.2. Performance Analysis
This section presents the comparative analysis of CSP
and after implementing our method in the Cloud environment.
Traffic at Datacenter
The traffic at datacenter includes the requests from legitimate users as well as attackers. This wil
contribute to flooding. Credit based
reputed users where as the well-
3, the users 2 & 6 have submitted 500 tasks per second and user 9 & 10 has submitted about
tasks per second. The users 3 & 5 have also submitted more than 100 requests per seconds.
Figure 3. Traffic at Datacenter before and after implementing solution
The increased spike, at the users 2, 6, 9
are suspicious users. The tasks submitted by these users are completely discarded without
disturbing the legitimate clients after employing our method. Thus,
the flooding at CSP and hence CSP can perform more efficiently even in the case of attack period.
Resource Utilization
Resource utilization here means how much percentage of CSP Datacenter resources are allotted to
each client. This includes the CPU, RAM and Bandwidth. As per
reputed users are given full access to CSP resources, reputed users
reputed users are fully blocked.
-100
0
100
200
300
400
500
600
USER1
USER2
Tasks
per
Second
As flow routers do the function of flow analysis and load balancer distributes the tasks to
routers which are not busy at the instant, there won’t be any flooding.
Notification mechanism to well reputed users about the likelihood of presence of
This section presents the comparative analysis of CSP’s performance in service delivery before
and after implementing our method in the Cloud environment.
The traffic at datacenter includes the requests from legitimate users as well as attackers. This wil
redit based system has completely eliminated the requests from ill
reputed users are given full access as before. As shown in Figure
, the users 2 & 6 have submitted 500 tasks per second and user 9 & 10 has submitted about
. Traffic at Datacenter before and after implementing solution
at the users 2, 6, 9 and 10, denoted that they are attackers and users 3
disturbing the legitimate clients after employing our method. Thus, credit based method reduced
and hence CSP can perform more efficiently even in the case of attack period.
each client. This includes the CPU, RAM and Bandwidth. As per the proposed method only well
reputed users are given full access to CSP resources, reputed users are given limited access and ill
USER2
USER3
USER4
USER5
USER6
USER7
USER8
USER9
USER10
USER11
USER12
Before
After
135
As flow routers do the function of flow analysis and load balancer distributes the tasks to
to well reputed users about the likelihood of presence of
’s performance in service delivery before
The traffic at datacenter includes the requests from legitimate users as well as attackers. This will
system has completely eliminated the requests from ill-
as before. As shown in Figure
, the users 2 & 6 have submitted 500 tasks per second and user 9 & 10 has submitted about 250
denoted that they are attackers and users 3 and 6
method reduced
and hence CSP can perform more efficiently even in the case of attack period.
method only well-
are given limited access and ill

136
Figure 4. Resource Utilization based on credits
The graph in Figure 4 depicts that well reputed users such as users 1, 4, 7,8,11 & 12 are given full
access to CSP resources. The users 3 & 5 who are suspicious are given limited access to resources
whereas the users 2, 6, 9 & 10 are completely blocked from accessing the CSP resources. Earlier
the users were given random resource allocation due to which well reputed users also faced
inefficient service delivery from CSP.
Processing Cost
The processing cost here means the cost incurred at each Datacenter in processing the requests
from all users.
Figure 5. Processing Cost per Datacenter
The processing cost at each datacenter has decreased tremendously after our method has been
applied. Instead of giving as much task as possible to one datacenter, the load is distributed
among the datacenters which will in turn lessen the response time for serving clients requests. As
shown in Figure 5, earlier only Datacenters1 & 2 does all the processing and other DCs were idle.
But, after implementing this credit based solution, all datacenters contributed to CSP service
delivery and hence helped in enhanced performance and reduced response time.

137
4. CONCLUSION
Credit based methodology could detect the DDoS attacks and discriminate it from the impatient
users i.e. flash crowds. This method could also prevent the environment from future attacks. The
experimental results proved the claim. . DDoS attacks are reported to be the number one threat
which risks the cloud service providers as well as customers with huge financial and reputation
loss. Credit based could achieve better resource utilization with reduction in cost. Hence this
method could be cost effective also. Cloud computing which has invaded almost the entire of IT
world is facing terrific setbacks due to various security issues prevailing in the cloud
environment. This method helps to mitigate the DDoS attack and at the same time processes the
flash crowd and provides them with requested service. The method is efficient in terms of
computational overhead and memory consumption. The communication between the entities
consumes time. Even though, owing to the adeptness of our method to detect and put off the
outrage of DDoS in cloud which handles critical business of and provide services to a huge
community, the communication overhead which may crop up can be ignored.
REFERENCES
[1] Haiqin Liu, Yan Sun, Victor C. Valgenti, and Min Sik Kim : TrustGuard: A Flow-level
Reputation-based DDoS Defense System. In: 5th IEEE Workshop on Personalized Networks,
pp.287-289 (2011)
[2] Maitreya Natu, Jelena Mirkovic: Fine-Grained Capabilities for Flooding DDoS Defense Using
Client Reputations. In: ACM, pp. 105-112 (2007)
[3] Ping Du, Akihiro Nakao: OverCourt: DDoS Mitigation through Credit-Based Traffic Segregation
and Path Migration. In: Computer Communications 33, pp. 2164–2175(2010)
[4] Shui Yu, Theerasak Thapngam, Jianwen Liu, Su Wei and Wanlei Zhou: Discriminating DDoS
Flows from Flash Crowds Using Information Distance. In: Third International Conference on
Network and System Security, IEEE, pp.351-356 (2009)
[5] Shui Yu, Wanlei Zhou, Weijia Jia, Song Guo, Yong Xiang, and Feilong Tang: Discriminating
DDoS Attacks from Flash Crowds Using Flow Correlation Coefficient. In: IEEE Transactions On
Parallel And Distributed Systems,, pp. 1-7 (2012)
[6] Theerasak Thapngam, Shui Yu, Wanlei Zhou and Gleb Beliakov. In: Discriminating DDoS Attack
Traffic from Flash Crowd through Packet Arrival Patterns. In: IEEE Conference on Computer
Communications Workshops, pp. 952– 957 (2011)
[7] QI Chen, Wenmin Lin, Wanchun Dou, Shui Yu: CBF- a packet filtering method for DDoS attack
defence in cloud computing. In: IEEE 9th International conference on Dependable, Autonomic and
Secure Computing,427-434, (2011)
[8] Ahmad Rashidi and Naser Movahhedinia: A Model for User Trust in Cloud Computing. In:
International Journal on Cloud Computing: Services and Architecture (IJCCSA), vol.2, No. 2
(2012)
[9] Priyank Singh Hada, Ranjita and Mukul Manmohan: Security Agents: A Mobile Agent based
Trust Model for Cloud Computing. In: International Journal of Computer Applications, (0975-
8887), vol.36 (2011)
[10] Wenjuan Li, Lingdi Ping and Xuezeng Pan: User Trust Management Module to Achieve Effective
Security Mechanisms in Cloud Environment. In: ICEIE,Vol. 1 (2010)

138
[11] Pritesh Jain,Dheeraj rane and Shyam patidar: A Novel Cloud Bursting Brokerage and
Aggregation(CBBA) Algorithm for Multi Cloud Environment. In: Second International
Conference on Advanced Computing &Communication Technologies (2012)
[12] Xiaodong Sun ,Guiran Chang and Fengyun Li: A Trust Management Model to enhance security of
Cloud computing Environment. In: 2nd International Conference on Networking and Distributed
Computing (2011)
[13] Mahbub Ahmed and Yang Xiang: Trust Ticket Deployment: A Notion of a Data Owners Trust in
Cloud Computing. In: International Joint Conference of IEEE TrustCom-11/IEEE ICESS-
11/FCST-11(2011)
[14] N. Jeyanthi and N.Ch.S.N.Iyengar, Packet Resonance Strategy: A Spoof Attack Detection and
Prevention Mechanism in Cloud Computing Environment, International Journal of
Communication Networks and Information Security, Vol.4,No.3, December 2012, pp.163-173.
[15] Jeyanthi, N.; Vinithra, J.; Sneha, S.; Thandeeswaran, R.; Iyengar, N.Ch.S.N., “A Recurrence
Quantification Analytical Approach to Detect DDoS Attacks”, IEEE International Conference on
Computational Intelligence and Communication Networks (CICN), 2011, pp.58 – 62.

CREDIT BASED METHODOLOGY TO DETECT AND DISCRIMINATE DDOS ATTACK FROM FLASH CROWD IN A CLOUD COMPUTING ENVIRONMENT

More Related Content

What's hot (17)

Similar to CREDIT BASED METHODOLOGY TO DETECT AND DISCRIMINATE DDOS ATTACK FROM FLASH CROWD IN A CLOUD COMPUTING ENVIRONMENT (20)

Recently uploaded (20)

CREDIT BASED METHODOLOGY TO DETECT AND DISCRIMINATE DDOS ATTACK FROM FLASH CROWD IN A CLOUD COMPUTING ENVIRONMENT